Rotarod Test: Protocol, Scoring, Reporting

The rotarod is one of the most-run motor assays in rodent neuroscience and one of the least standardised. Two labs can both write “accelerating rotarod, 4–40 rpm, latency to fall” in their methods and still be running measurably different experiments, because rod diameter, surface texture, training schedule, endpoint definition and the drum’s actual (not nominal) acceleration all move the number. Crabbe and colleagues demonstrated the consequence directly in PNAS: exactly how the rotarod test is performed can markedly alter the apparent patterns of genetic influence, and the genetic contribution to accelerating versus fixed-speed performance can be completely dissociated under some test conditions. Choose the parameters badly and you do not get a noisy answer — you get a confident answer to a different question.

That is the practical framing for this page: the rotarod protocol is not a preamble to the assay, it is the assay. Everything below is drawn from published SOPs and method papers, and where the literature disagrees, it says so rather than inventing a consensus value.

Decide first: are you measuring coordination or motor learning?

These are different endpoints and they call for different designs, yet the same phrase — “rotarod performance” — is used for both.

Baseline coordination and balance. A single session with no prior training. The International Mouse Phenotyping Consortium (IMPC) SOP takes this approach: no training period before testing, mice acclimated in the room in their home cages for at least 15 minutes, then three trials separated by 15-minute inter-trial intervals.
Motor learning. Repeated sessions over consecutive days, where the outcome of interest is the slope of improvement rather than day-one performance. Published multi-day designs run three trials per day for five consecutive days, or four trials per day across four days.

The apparatus should follow the question. In a 2023 eNeuro study that compared rod geometries head to head, the small rod (4.445 cm diameter) gave the greatest sensitivity to initial motor performance, while the large rod (10.16 cm) gave the greatest sensitivity for motor learning; laddered rods presented the greatest challenge and produced the lowest latencies. If you run a small smooth rod and then report a learning curve, you have chosen the geometry least suited to the claim.

The apparatus is a protocol variable, not a background detail

Rod diameter varies roughly three-fold across setups in the published literature, and it is not interchangeable:

A widely used commercial mouse rotarod (Ugo Basile 47650) specifies a 3 cm rod, five 5.7 cm lanes, a 16 cm fall height to the trip box, speeds adjustable from 3 to 80 rpm in 1 rpm steps, and constant, ramp, reverse-ramp, custom-ramp and rocking modes. Falls are detected by a magnetic trip-box sensor that logs elapsed time, revolutions, distance and speed per lane.
The IMPC SOP describes a rod of approximately 5 cm diameter, hard plastic covered with grey rubber foam, with lanes about 5 cm wide.
Deacon’s JoVE protocol uses a 3 cm knurled rod with parallel ridges, 30 cm above the base, with 30 cm flanges set 6 cm apart, and stresses that rod diameter and ridge depth critically affect grip quality.

The eNeuro comparison explains why this matters mechanistically: smaller-diameter drums can encourage passive behaviour, because an animal can cling to a thin rod and be carried round, while larger-diameter rods recruit walking and running ability in addition to balance and coordination. A change in rod diameter therefore changes the behavioural construct, not just the scale of the numbers. Never pool latencies across apparatus, and never compare your absolute latencies to a published figure obtained on a different rod.

Calibrate the acceleration — do not trust the dial

This is the single most overlooked source of between-lab discrepancy. A Journal of Neuroscience Methods paper on rotarod calibration makes the point plainly: latency to fall can differ markedly between laboratories using the same brand of rod, and while that can arise from diameter, texture, protocol or environment, it is also possible that the actual acceleration rates of the devices do not correspond to the nominal rates set on them. The paper describes a simple method to measure the acceleration rate of any rotarod and set it to a desired value.

Practically: measure the real ramp before a study starts, after any service or belt replacement, and periodically during long studies; log the measurement alongside your other equipment records. A drum that ramps 10 % faster than the display claims will shorten every latency in the study, and no amount of statistics recovers that.

Training schedules that appear in the literature

There is no single canonical schedule. Treat the following as a menu of validated designs and pick — and then pre-specify — one, rather than assembling a hybrid after seeing the data.

IMPC phenotyping pipeline: no training; at least 15 minutes of room acclimation in the home cage; mice placed on the rod at a constant 4 rpm; acceleration from 4 to 40 rpm over 300 s; three trials with 15-minute inter-trial intervals; testing at approximately the same time each day.
Deacon (JoVE): start at 4 rpm, accelerate at 20 rpm/min to a 40 rpm maximum; begin acceleration 10 s after placement and only once the mouse is facing forward.
Two-day design (Tg4-42 Alzheimer’s model study): four trials per day on two consecutive days, inter-trial intervals of 10–15 minutes, 4 to 40 rpm over a 300 s trial.
Five-day design (eNeuro): acceleration from 4 to 60 rpm, three trials per day with at least 5 minutes between trials, five consecutive days, with each trial ending on a fall, two rod revolutions, or 3 minutes elapsed.
Four-day training plus modified test (BMC Biology, 2023): four consecutive training days at four trials per day, then a modified 5-minute test.

One finding from that 2023 BMC Biology study is worth building into any schedule: on the fourth training day, trials two, three and four were more consistent than trial one — probably reflecting habituation or learning — and there was no difference among trials two, three and four, leading the authors to conclude that two trials may be sufficient. In other words, trial one is substantially a habituation trial. Decide in advance whether you drop it; deciding afterwards is a garden-of-forking-paths problem, one of the mechanisms that drove the wider replication crisis.

Scoring: define the endpoint before the first animal goes on the rod

A trial can end in at least three distinguishable ways, and only one of them is an unambiguous fall.

Fall. The animal loses grip and drops to the trip box.
Passive rotation. The animal clings and is carried around the rod without walking. The IMPC SOP stops the timer when a mouse completes one full passive rotation; the eNeuro protocol ends the trial after two rod revolutions. Different thresholds, both explicit — and that explicitness is the point. If you do not define and record passive rotation, a mouse that grips and spins is scored as competent, which systematically inflates the performance of exactly the animals most likely to be impaired.
Jumping off. The IMPC SOP records the reason a trial ended — falling, jumping or passive rotation — as a separate data field alongside latency.

Which number to record

On an accelerating protocol, latency to fall and speed at fall carry the same information only if the ramp is identical; speed at fall is the more transferable figure when ramps differ between studies. Deacon’s protocol adds two rules that are easy to omit and hard to reconstruct later: a mouse that falls before the 10-second mark is replaced and retried, up to three attempts in total, with the first fall after 10 s recorded; and the datum is the mean speed across attempts, not the maximum — a mouse falling at 4 rpm and then 12 rpm scores 8 rpm, and a mouse that fails to grip within 10 s on all three attempts is assigned 4 rpm.

Watch the ceiling

With a 300 s cutoff, ceiling effects are real, not theoretical: in the eNeuro rod comparison, most mice on the smaller rods reached the 300 s cutoff by testing day 5, and the large unladdered rod showed the highest statistical power. If your control group is sitting at the cutoff, the assay has no headroom to detect improvement and limited headroom to detect impairment. The fixes are to extend the cutoff, raise the top speed, or use a more demanding rod geometry — all decisions to make before the study, not after.

Beyond latency: multi-parameter scoring

The 2023 BMC Biology group modified the trial structure by putting mice back on the rod after each fall until a fixed 5-minute trial ended, and added three read-outs to latency: longest duration on the rod (s), maximal distance covered (cm), and number of falls. Applied to mice with mild-to-moderate traumatic brain injury, normalising each animal’s data to its own baseline was needed to reduce individual variability, four trials were more sensitive than two, and maximal distance was the best parameter for detecting statistically significant long-term motor deficits. If your model produces subtle deficits and latency alone is coming back null, this is the most evidence-backed alternative in the current literature.

Confounds to pre-empt

Body weight. In Collaborative Cross mice, speed at fall correlated significantly with body weight in both sexes (females r = −0.56, p = 2.17E−15; males r = −0.53, p = 1.89E−15). The literature is not unanimous — correlations have been reported in some inbred strains and absent in others — so the defensible practice is to record body weight at the time of testing, as the IMPC SOP does, and either include it as a covariate or report the observed correlation in your sample. Do not assume it away, and do not assume it applies.
Dose–response is not monotone for some drugs. Rustay and colleagues found that moderate to high ethanol doses disrupt rotarod performance while under certain conditions low doses enhanced accelerating rotarod performance — which is why at least two doses are needed to characterise individual sensitivity differences, and why a single-dose null result is uninformative.
Fatigue. Deacon argues a mouse is unlikely to be markedly fatigued after just 2–3 minutes on the rod. That reasoning does not automatically extend to 5-minute trials repeated four times a day, so shorter inter-trial intervals demand more justification, not less.
Olfactory and environmental cues. The IMPC SOP cleans the apparatus between trials with water and then 50 % ethanol, and tests at approximately the same time each day to control for physiological variation. Handling history and housing conditions also shape rodent behaviour, so keep husbandry and handling constant across groups and describe them.
Downstream contamination of other assays. Crabbe and colleagues note that impaired motor performance can confound the interpretation of learning and memory, exploration, motivation and sensory-competence assays. Establish motor status before you interpret maze or operant data from the same cohort.

The compliance and reporting layer

Rotarod testing is usually low-severity, but “low severity” is a conclusion the committee reaches, not an assumption you make. Your IACUC protocol should describe the fall height, the number and duration of trials, the handling and placement method, the endpoint criteria including passive rotation, cleaning between animals, and what happens to an animal that repeatedly fails to grip or shows distress. Where the same cohort receives drugs or surgery, the rotarod schedule needs to be reconciled with recovery and dosing timelines.

For reporting, the relevant benchmark is ARRIVE 2.0, published with its Explanation and Elaboration in PLOS Biology in 2019. Its Essential 10 are: study design; sample size; inclusion and exclusion criteria; randomisation; blinding/masking; outcome measures; statistical methods; experimental animals; experimental procedures; and results. A further eleven items (11–21) form the Recommended Set. Three of the Essential 10 are routinely thin in rotarod papers — inclusion/exclusion criteria (what happened to the animals that jumped or never gripped?), randomisation (lane and running order, not just group assignment), and blinding (the person placing the animal usually knows the genotype).

For NIH-funded work, the rigor and transparency expectations cover scientific premise, rigorous experimental design, consideration of relevant biological variables, and authentication of key resources. NIH defines scientific rigor as “the strict application of the scientific method to ensure robust and unbiased experimental design, methodology, analysis, interpretation and reporting,” and instructs applicants to discuss their consideration of sex as a biological variable in vertebrate animal studies. A rotarod cohort of one sex, or an analysis that pools sexes without testing for an interaction, is a reviewable gap — and since body weight differs by sex, the weight recording described above does double duty as a methodological and a rigor-reporting control.

A rotarod methods paragraph that survives review

Apparatus make and model, rod diameter, surface material and texture, lane width, fall height, and how a fall is detected and timed.
Measured (not nominal) acceleration profile: start speed, ramp rate or ramp duration, top speed, and when the calibration was performed.
Training schedule: number of days, trials per day, inter-trial interval, and whether any trial was excluded from analysis.
Placement method and orientation rule, and any pre-cutoff re-placement rule.
Endpoint definitions for fall, jump and passive rotation, and the revolution threshold used for the latter.
Trial cutoff, how ceiling values were handled statistically, and the proportion of animals reaching cutoff.
Read-outs reported: latency, speed at fall, distance, number of falls, or a normalised composite.
Body weight at test, sex and age of every group; time of day of testing; cleaning procedure.
Randomisation of lane and run order, who was blinded, and the experimental unit used in the analysis.

Statistical handling

The animal is the experimental unit; trials within an animal and days within an animal are repeated measures, not independent observations. The Tg4-42 study analysed rotarod performance with two-way repeated-measures ANOVA and Bonferroni post-hoc comparisons, which is a conventional and defensible choice for a balanced multi-day design.

Two decisions deserve to be pre-specified rather than improvised. First, latency data with a fixed cutoff are censored at the cutoff — treating a 300 s value as an observed latency understates the true performance of the animals that never fell, and if a substantial fraction of a group is at cutoff, a method that acknowledges censoring, or a harder task, is more honest than a mean. Second, when within-animal variability dominates, normalising each animal to its own pre-treatment baseline is well supported: the BMC Biology group found normalisation to individual baseline necessary to reduce individual differences before deficits became detectable. Decide both before unblinding.

Frequently asked questions

Do mice need training days before rotarod testing?

It depends on what you are measuring. The IMPC pipeline runs no training period at all — three trials on a single day with 15-minute intervals — because the target is baseline coordination in a high-throughput screen. If your outcome is motor learning, multi-day designs of three to four trials per day over four to five days are standard in the literature. A design with training days and a single-day analysis is measuring something in between; say which one you mean.

Should I report latency to fall or speed at fall?

Report both if your equipment logs both. On an accelerating ramp they encode the same event, but speed at fall is comparable across studies with different ramp rates, whereas raw latency is not. Deacon’s protocol treats speed at fall as the primary datum and takes the mean across attempts rather than the maximum.

How many trials per day, and how long between them?

Published inter-trial intervals range from at least 5 minutes to 15 minutes, and trials per day from three to four. The IMPC uses three trials at 15-minute intervals; a five-day eNeuro protocol used three trials at a minimum of 5 minutes. The 2023 BMC Biology analysis found trials two through four equally consistent and concluded two trials may suffice, with trial one behaving as habituation.

Does body weight invalidate rotarod results?

Not automatically, but it can confound a group comparison when groups differ in weight — and in Collaborative Cross mice the correlation between body weight and speed at fall was substantial in both sexes. Other studies in other strains have found no such correlation. Record weight at testing, report the correlation in your own sample, and treat weight as a covariate when groups are unbalanced.

What counts as a fall if the mouse clings and rotates instead of walking?

That is a passive rotation, and it must be defined in advance because instruments generally will not detect it. The IMPC stops the timer at one complete passive rotation and records the reason for trial end; another published protocol ends the trial at two rod revolutions. Thin rods make passive clinging more likely, so a small-diameter drum makes this rule more consequential, not less.

Is the rotarod sensitive enough to detect mild deficits?

Latency alone often is not. The evidence-backed route is to add read-outs rather than add animals: longest duration on the rod, maximal distance covered and number of falls over a fixed-length trial with re-placement after each fall, with each animal normalised to its own baseline. In a mild-to-moderate traumatic brain injury model, maximal distance was the parameter that detected long-term deficits when latency did not.

Can I compare my latencies to published values?

Only with a matched apparatus and a matched, calibrated ramp. Rod diameters in the published literature range from 3 cm to over 10 cm, surfaces range from knurled plastic to rubber foam to laddered designs, and measured acceleration may not match the value set on the device. Absolute latencies are properties of a rig and a protocol; only within-study group differences travel.