ATR-FTIR vs Transmission FTIR: Penetration Depth, Crystal Choice, and Spectral Distortion

Nearly every routine mid-infrared measurement now starts on an ATR accessory, because it needs no sample preparation. That convenience hides a real decision: attenuated total reflectance and transmission do not sample the same amount of material, do not sample the same amount of material at every wavenumber, and do not produce the same band positions. This guide covers the arithmetic behind the sampling depth, how to pick a crystal from the sample’s chemistry rather than from habit, and the specific distortions ATR introduces relative to transmission — which is the reason an ATR spectrum searched against a transmission library can return the wrong answer.

For reading the resulting spectrum — band assignment, the fingerprint region, worked interpretation — see the companion guide on IR functional group frequencies and how to read a spectrum. This page is about choosing the sampling mode and understanding what that choice does to the data.

The decision, stated plainly

Use ATR when you need speed, when the sample is intact and you want it back, when the material is opaque, strongly absorbing, aqueous, a paste, a film or a surface, or when the analytical question is a surface or top-few-microns question. Use transmission when the answer must be quantitative against a Beer–Lambert calibration, when you need band positions and relative intensities that match published transmission reference data without post-processing, when the analyte is present at low concentration in a bulk matrix, or when the sample cannot be brought into intimate optical contact with a crystal.

The failure modes are asymmetric. Bad ATR usually produces a spectrum that looks plausible and is quietly wrong — weak because contact was poor, sloped because the penetration depth changed across the spectrum, or shifted because of dispersion near a strong band. Bad transmission usually produces a spectrum that looks obviously bad: saturated bands, a hydrated potassium bromide pellet, mineral-oil interference. A wrong-looking spectrum gets rerun. A plausible-looking wrong spectrum gets reported.

How deep does ATR actually sample?

There is no single depth of penetration for ATR, and any source quoting “about one micron” without conditions has dropped three of the four variables. The standard expression, from the internal-reflection theory that ASTM E573 summarises, is:

d_p = λ ÷ ( 2π n₁ √( sin²θ − (n₂/n₁)² ) )

with the variables defined as:

• d_p — the penetration depth: the distance from the crystal–sample interface at which the evanescent field has decayed to 1/e (about 37%) of its value at the interface. It is not the depth to which the sample is fully interrogated, and it is not an effective pathlength.
• λ — the free-space wavelength of the radiation, in the same units you want d_p in. Because λ = 10,000 / (wavenumber in cm⁻¹) gives micrometres, 1,000 cm⁻¹ is 10 µm and 4,000 cm⁻¹ is 2.5 µm.
• n₁ — the refractive index of the internal reflection element (the crystal).
• n₂ — the refractive index of the sample.
• θ — the angle of incidence at the crystal–sample interface, measured from the normal. Most commercial single-reflection accessories are fixed at 45°.

The assumptions the equation carries

Quoting a d_p number without these is where most sampling tables go wrong:

• θ exceeds the critical angle. Below it there is no total internal reflection and the equation does not apply at all. See the next section — this is a live constraint, not a formality.
• The sample is optically semi-infinite relative to d_p. A film thinner than a few d_p is sampled through to its substrate, and the substrate contributes bands.
• The sample is weakly absorbing and n₂ is real and constant. This is the assumption that fails hardest, and it fails precisely where the information is: inside a strong absorption band, n₂ is neither constant nor purely real. That is the anomalous-dispersion problem discussed below.
• Optical contact is intimate. The evanescent field decays over micrometres; an air gap of a micrometre is not a small error, it is most of the sampling depth.
• A single, plane-wave reflection. Real accessories have a distribution of angles and, in multi-reflection designs, many bounces — which multiplies the effective pathlength without changing d_p.

Worked example

A diamond element (n₁ = 2.4) at 45°, against a typical organic solid (n₂ ≈ 1.5), at 1,000 cm⁻¹:

• λ = 10,000 / 1,000 = 10 µm
• (n₂/n₁)² = (1.5 / 2.4)² = 0.625² = 0.3906
• sin²45° = 0.5, so the bracket is 0.5 − 0.3906 = 0.1094, and its square root is 0.3308
• d_p = 10 ÷ (2π × 2.4 × 0.3308) = 10 ÷ 4.99 = 2.0 µm

That result matches the 2 µm Specac publishes for a standard diamond puck under exactly those stated conditions (1,000 cm⁻¹, 45°, sample refractive index 1.5), which is a useful check that you have the equation the right way round.

d_p is not constant across the spectrum — and that is the whole problem

Because d_p is directly proportional to λ, the sampling depth on that same diamond element varies across a single scan:

Wavenumber (cm−1)	λ (µm)	dp, diamond n1=2.4 (µm)	dp, Ge n1=4.0 (µm)
4,000 (O–H, N–H stretch)	2.5	0.50	0.17
3,000 (C–H stretch)	3.33	0.67	0.22
1,700 (C=O stretch)	5.88	1.18	0.39
1,000 (fingerprint)	10.0	2.00	0.66
650 (low cut-off)	15.4	3.08	1.02

Computed from the equation above at 45° with n₂ = 1.5. The sampling depth at 650 cm⁻¹ is roughly six times that at 4,000 cm⁻¹. The same sample, in the same scan, is being interrogated over six times more material at the low-wavenumber end than at the high end. Nothing about the chemistry changes; the optics change.

d_p is not an effective pathlength

Do not substitute d_p into Beer–Lambert as a pathlength. The effective pathlength depends additionally on the electric-field amplitudes at the interface, the polarisation, and the number of active reflections in the accessory. Specac’s own table, for instance, pairs a 2 µm d_p for diamond with a quoted effective pathlength of 4.36 µm, and a 2 µm d_p for zinc selenide with 2.83 µm — same penetration depth, different pathlength. If you are building a quantitative ATR method, take the effective pathlength from your accessory’s own documentation, keep every geometric variable fixed, and calibrate empirically. For how detection limits are then established from that calibration, see calculating LOD and LOQ.

The constraint most sampling tables omit: the critical angle

Total internal reflection only occurs above the critical angle, θ_c = arcsin(n₂/n₁). Rearranged for a fixed-45° accessory, this sets a hard ceiling on the sample refractive index you can measure at all:

n₂ < n₁ × sin 45° = n₁ × 0.7071

Crystal	n1	θc for n2 = 1.5	Max sample n2 at 45°
Diamond	2.4	38.7°	1.70
Zinc selenide (ZnSe)	2.41	38.5°	1.70
Silicon (Si)	3.41	26.1°	2.41
Germanium (Ge)	4.00	22.0°	2.83

Two practical consequences. First, on diamond or ZnSe at 45° the margin above the critical angle for an ordinary organic sample is only about six degrees — which is why anything that raises the local refractive index (a high-index filler, carbon black, a strongly dispersive band) can push the measurement toward or through the critical angle and produce badly distorted, derivative-shaped bands. Second, high-refractive-index materials such as heavily filled elastomers, carbon-loaded polymers and many inorganic pigments genuinely cannot be measured on a 45° diamond element, and germanium is not a stylistic preference for those samples but a requirement.

Choosing the crystal from the sample’s chemistry

Crystal	n1	Relative dp	Chemical limits	Mechanical notes	Best for
Diamond	2.4	Highest (with ZnSe)	Chemically inert across the full pH range; the default for acids, bases and aggressive solvents	Hardest available; tolerates high clamp pressure on hard solids and abrasive powders	General-purpose routine work; anything corrosive; hard or gritty solids
Zinc selenide	2.41	Highest (with diamond)	Restricted; Specac states samples should be kept within pH 5–9. Attacked by acids — and acid attack on a selenide liberates hydrogen selenide, which is acutely toxic. Also attacked by strong bases and some oxidisers	Soft (Mohs ≈ 2.5); scratches easily, so unsuitable for hard powders under pressure	Neutral liquids, solvents, soft films; longer optical paths in multi-reflection accessories
Germanium	4.00	Lowest — roughly a third of diamond	Good acid resistance; damaged by strong bases	Harder than ZnSe but brittle; becomes optically opaque at elevated temperature (free-carrier absorption), so it is a poor choice for heated stages	High-refractive-index samples (carbon-filled rubbers, dark polymers), strongly absorbing samples, genuine surface and thin-coating analysis
Silicon	3.41	Low	Broadly resistant to acids and to many bases	Hard and robust; thin elements are fragile in handling	Aggressive chemistry needing a lower dp than diamond; far-infrared work — but see the transmission gap below

Where the vendors genuinely disagree

Published crystal transmission ranges differ substantially between manufacturers, and the disagreement is real rather than typographical. Specac’s technical material quotes ZnSe at 7,800–500 cm⁻¹, germanium at 5,500–480 cm⁻¹, silicon at 8,000–1,350 plus 500–33 cm⁻¹, and a standard diamond puck at 7,800–400 cm⁻¹ with an extended version reaching 10,000–10 cm⁻¹. Other instrument vendors’ summary tables commonly quote much narrower working ranges for the same materials — ZnSe at roughly 4,000–650 cm⁻¹ and germanium at roughly 4,000–870 cm⁻¹ are both widely published.

Both sets of figures can be correct, because a “range” depends on the element’s thickness, whether it is anti-reflection coated, whether the figure describes the crystal alone or the assembled accessory including its own optics and detector, and on the transmittance threshold the vendor chose to call a cut-off. The operational rule: do not take a cut-off from a generic comparison table — including this one. Take it from the technical note for the specific accessory and element you own, and if a band near your cut-off is load-bearing for the identification, confirm it against a reference material on that same accessory.

Two further material-specific points worth checking against your own vendor’s note rather than assuming. Diamond has intrinsic two-phonon lattice absorption in the mid-infrared — commonly cited as roughly the 2,650–1,900 cm⁻¹ region — whose severity scales with the thickness of the diamond element, so noise rises in that window on some designs and is negligible on others. And silicon’s transmission gap between about 1,350 and 500 cm⁻¹ in Specac’s figures removes most of the fingerprint region, which disqualifies it as a general-purpose identification crystal even though its chemical resistance is attractive. Thallium bromoiodide (KRS-5) is still encountered on older far-infrared accessories; it was not covered by the vendor documentation checked for this guide, and it is soft, slightly water-soluble and toxic, so treat any KRS-5 element as a specialist case and consult its own documentation.

What ATR does to the spectrum that transmission does not

This is the section that matters most and is most often skipped. Three separate distortions operate, and they are not the same distortion.

1. The intensity slope: relative band intensities rise toward lower wavenumber

Since d_p scales with λ, more material is sampled at low wavenumber than at high. The direct consequence, and it is visible on essentially every uncorrected ATR spectrum: fingerprint-region bands are relatively stronger, and high-wavenumber bands such as O–H and N–H stretches relatively weaker, than the same bands in a transmission spectrum of the same material. Miseo and Larkin state this directly in Applied Spectroscopy Practica (2025): higher-frequency bands such as the O–H stretch are less intense while lower-frequency fingerprint bands absorb more strongly.

The practical trap is diagnostic rather than cosmetic. An analyst comparing an uncorrected ATR spectrum against a remembered transmission spectrum will systematically under-read hydroxyl and amine content and over-read fingerprint features. This is not a concentration difference and no amount of resampling will remove it.

2. Band shifts from anomalous dispersion

The penetration-depth equation assumes n₂ is constant. It is not. By the Kramers–Kronig relationship, the real refractive index of a material varies sharply across an absorption band — rising on the low-wavenumber side and falling on the high-wavenumber side. Because d_p depends on n₂, this varying index modulates the sampling depth within the band, producing an asymmetric, partly derivative-shaped (“bisignate”) band profile and shifting the apparent maximum to lower wavenumber relative to transmission.

Miseo and Larkin give a concrete measured case: the carbonyl band of poly(methyl methacrylate) appears at 1,731 cm⁻¹ in transmission and at 1,724 cm⁻¹ on a diamond ATR — a 7 cm⁻¹ shift that they describe as substantial and as sufficient to cause misidentification if search results are not interpreted critically. Seven wavenumbers is comfortably enough to move a carbonyl assignment from one ester environment to another. The effect is largest for the strongest bands, which are exactly the bands an analyst leans on.

3. The ATR correction — and what the simple version does not fix

Instrument software offers an ATR correction, and the important thing to know is that there are two classes of it:

• A simple correction rescales intensities for the wavenumber dependence of d_p. It fixes distortion 1. It does not fix band shifts.
• An advanced correction addresses both the wavenumber intensity dependence and the peak shifts arising from anomalous dispersion.

Miseo and Larkin further note that the advanced correction works reliably only when the refractive index difference between crystal and sample (Δn) exceeds roughly 1 — which is a real constraint, since a diamond or ZnSe crystal at n ≈ 2.4 against an organic sample at n ≈ 1.5 gives Δn ≈ 0.9, just below that. Two operational habits follow. Know which correction your software actually applied, because the label “ATR corrected” does not distinguish them. And record whether a spectrum is corrected or not in the file metadata, because a corrected and an uncorrected spectrum of the same sample are different data and are not interchangeable downstream.

4. Why library matching fails, and what to do about it

Put the three effects together and the reason an ATR spectrum searched against a transmission library returns a poor or wrong hit is fully explained: the relative intensity envelope is tilted, the strongest bands are shifted by several wavenumbers, and their shapes are asymmetric. A curve-matching algorithm sees all three as chemical differences.

The mitigations, in order of preference:

1. Search an ATR library with ATR data. Commercial libraries are published in both forms; matching like with like removes the problem at source rather than compensating for it.
2. If you must search a transmission library, apply an advanced ATR correction first and record that you did, subject to the Δn caveat above.
3. Treat the hit-quality index as a screening number, not a conclusion. ASTM E2310 is the standard guide covering the use of spectral searching by curve-matching algorithms with mid-infrared data; the discipline it describes — inspecting the overlay rather than accepting the rank order — is exactly what catches an ATR-versus-transmission mismatch.
4. Confirm identification against a reference material measured on your own accessory, in the same sampling mode, at the same pressure. In regulated or evidential work this is not optional; it is the difference between a library hit and an identification. Forensic identification workflows treat this explicitly — see forensic drug analysis standards and instrumentation for how confirmatory identification is structured there.

When transmission is still the right answer

• Quantitation against a Beer–Lambert calibration. A fixed-pathlength liquid cell has a known, constant pathlength. An ATR measurement’s effective pathlength depends on contact pressure, sample refractive index and wavenumber — three variables you would rather not have in a calibration.
• Trace analyte in a bulk matrix. ATR interrogates micrometres; transmission interrogates the whole pathlength. For a dilute component, transmission simply sees more of it.
• Comparability with published transmission reference data where applying and documenting a correction is more trouble than pressing a pellet.
• Samples that will not make optical contact — rigid, rough, curved or brittle solids that cannot be deformed against a crystal. Grinding into a halide pellet or a mineral-oil mull sidesteps the contact problem entirely.
• Hard powders on a soft crystal. If diamond is unavailable, the pressure needed for contact may damage a ZnSe element; a pellet is cheaper than a crystal.

The classic transmission trade-offs still apply: potassium bromide is hygroscopic and residual water introduces a broad O–H band that can be mistaken for the analyte’s own, and mineral-oil mulls contribute strong C–H bands that obscure the sample’s C–H region.

A selection checklist

1. Is the question a surface question or a bulk question? Surface, coating or thin-film → ATR, and consider germanium for the shallowest d_p. Bulk composition or trace analyte → transmission.
2. Estimate the sample’s refractive index. Above about 1.7, a 45° diamond or ZnSe element will not sustain total internal reflection — move to germanium.
3. Check the chemistry against the crystal. Acidic or strongly basic → diamond. Never acid on ZnSe: pH 5–9 is the stated working window and acid attack liberates hydrogen selenide.
4. Check the mechanics. Hard or abrasive solid needing clamp pressure → diamond. Elevated temperature → not germanium.
5. Confirm the cut-off you need is inside your accessory’s real range, from its own technical note, not a generic table.
6. Decide the correction before you acquire, not after: which correction, applied or not, recorded in the metadata.
7. Match the library to the sampling mode. ATR data against an ATR library; otherwise correct first and inspect the overlay.
8. If the result is quantitative, fix every geometric variable — accessory, crystal, angle, pressure — and calibrate empirically rather than from d_p.

Frequently asked questions

What is the depth of penetration of ATR-FTIR?

There is no single value. It is given by d_p = λ / (2πn₁√(sin²θ − (n₂/n₁)²)) and therefore depends on wavenumber, the crystal’s refractive index, the sample’s refractive index and the angle of incidence. For a diamond crystal at 45° against an organic sample of index 1.5 it is about 2 µm at 1,000 cm⁻¹, but about 0.5 µm at 4,000 cm⁻¹ and about 3 µm at 650 cm⁻¹ — in the same scan. Any quoted figure is meaningless without its four conditions.

Why does my ATR spectrum not match the transmission reference spectrum?

Three effects, all optical rather than chemical. Relative intensities are tilted because sampling depth increases toward lower wavenumber. Strong bands shift to lower wavenumber through anomalous dispersion — measurably 7 cm⁻¹ for the PMMA carbonyl on diamond. And those bands become asymmetric in shape. Apply an advanced ATR correction, or better, compare against an ATR-acquired reference.

Which ATR crystal should I use for acidic samples?

Diamond. Zinc selenide should be kept to samples in roughly pH 5–9; acid attack on ZnSe liberates hydrogen selenide, which is acutely toxic, making this a safety constraint as well as an instrument-protection one. Germanium tolerates acids reasonably but is damaged by strong bases and is not suited to heated work.

Does the ATR correction in my software fix band positions?

Only if it is an advanced correction. A simple correction rescales intensity for the wavenumber dependence of d_p and leaves band positions where they are. The software label “ATR corrected” does not tell you which was applied — check the algorithm, and note that advanced corrections have been reported to work reliably only where the crystal–sample refractive index difference exceeds about 1.

Can I use ATR for quantitative analysis?

Yes, but not by treating d_p as a Beer–Lambert pathlength. The effective pathlength differs from d_p, depends on polarisation and the number of reflections, and varies with contact pressure and sample refractive index. Fix every geometric variable, calibrate empirically against standards measured on the same accessory, and validate the method rather than transferring a transmission calibration.

Why is my whole ATR spectrum weak?

Almost always optical contact rather than concentration. The evanescent field decays over a few micrometres, so a small air gap removes most of the signal. Increase clamp pressure within the crystal’s mechanical limit, grind or flatten the sample, or reconsider the sampling mode — a rigid, rough solid that will not deform is a transmission sample.

What does germanium buy me over diamond?

A lower penetration depth — roughly a third of diamond’s under the same conditions — and a much higher critical-angle margin, which is what allows high-refractive-index and strongly absorbing samples to be measured at all. The costs are a narrower spectral range, brittleness, vulnerability to strong bases, and loss of transparency at elevated temperature.

Standards and further reading

• ASTM E573, Standard Practices for Internal Reflection Spectroscopy — the governing practice for ATR, covering the underlying theory, the parameters that determine the result, and interpretation features specific to internal reflection. Current designation E573-01(2021), from ASTM subcommittee E13.03.
• ASTM E1252, Standard Practice for General Techniques for Obtaining Infrared Spectra for Qualitative Analysis — covers 4,000–50 cm⁻¹ across liquid, solid and vapour sampling. Current designation E1252-98(2021).
• ASTM E2310, Standard Guide for Use of Spectral Searching by Curve Matching Algorithms with Data Recorded Using Mid-Infrared Spectroscopy — the reference point for library-search discipline.
• ASTM E168, Standard Practices for General Techniques of Infrared Quantitative Analysis.
• E. V. Miseo and P. J. Larkin, “Fourier Transform Infrared Spectroscopy (FT-IR) Diamond Attenuated Total Reflection (ATR) Measurements: The Good, the Bad, and the (Really) Ugly,” Applied Spectroscopy Practica, vol. 3, no. 2, 2025 — the source for the anomalous-dispersion band-shift figures and the simple-versus-advanced correction distinction used above.
• Specac technical material on ATR crystal choice, for the comparative crystal table quoted at 1,000 cm⁻¹, 45° and sample refractive index 1.5.
• Crystal transmission ranges and chemical compatibility should always be confirmed against your own instrument vendor’s documentation — Thermo Fisher, PerkinElmer, Bruker, Agilent, Shimadzu and Specac all publish accessory-specific notes, and as shown above they do not all agree.

Related CASRAI guides

• IR spectroscopy: functional group frequency table and how to read a spectrum — band assignment once the sampling mode is settled.
• Atomic absorption spectroscopy: flame versus graphite furnace — the same shape of decision, made on sensitivity and matrix instead of penetration depth.
• Thermogravimetric analysis: running a scan and interpreting mass-loss steps — frequently paired with FTIR for evolved-gas identification.
• Limit of detection versus limit of quantitation — establishing the numbers a quantitative ATR method has to report.
• Dynamic light scattering: Z-average, PDI and reading a DLS report — another technique whose headline number hides its model assumptions.
• Laboratory equipment — the full cluster hub for instrument selection, operation and maintenance.