NumPy Array: Create from a List, Convert Back and Read Shape

A NumPy array is created with numpy.array(): pass it a list (or a list of lists) and it returns an ndarray, NumPy’s homogeneous multidimensional array. To go the other way, ndarray.tolist() returns a nested list of ordinary Python numbers. This guide covers both directions, shows how to read an array’s shape and type, and explains the one difference that trips people up: list(arr) and arr.tolist() do not return the same thing. The details come from the official NumPy documentation, linked under Sources.

What a NumPy array is used for in research

The NumPy quickstart describes the main object as “the homogeneous multidimensional array”: a table of elements, usually numbers, all of the same type, indexed by a tuple of non-negative integers. Turning a Python list into an array is typically the first step before the numerical work shown in guides such as ANOVA in Python with statsmodels and train test split in Python. For the broader practice of keeping such scripts repeatable, see computational reproducibility.

Install NumPy and import it

The NumPy install page gives two routes. With pip: pip install numpy. With conda, create and activate an environment, then install:

conda create -n my-env
conda activate my-env
conda install numpy

The conda route uses the defaults or conda-forge channel. If you want the environment to be portable, see sharing a reproducible conda environment. By convention NumPy is imported under the short name np, and you can confirm what is installed by printing the version:

import numpy as np
print(np.__version__)

Convert a Python list to a NumPy array (list to array numpy)

Pass the list as the single first argument to np.array(). The array type is deduced from the type of the elements in the sequence.

1. Import NumPy: import numpy as np.
2. Pass one list: np.array([1, 2, 3]) returns array([1, 2, 3]).
3. Mix integers and floats and NumPy promotes to a type that can hold every value: np.array([1, 2, 3.0]) returns array([ 1., 2., 3.]).
4. Pass a list of lists for two dimensions: np.array([[1, 2], [3, 4]]) returns a 2 by 2 array. The documentation states that array transforms sequences of sequences into two-dimensional arrays, sequences of sequences of sequences into three-dimensional arrays, and so on. Tuples inside a list work the same way: np.array([(1.5, 2, 3), (4, 5, 6)]) gives a 2 by 3 array of floats.

Parameters worth knowing

The signature in the reference page is numpy.array(object, dtype=None, *, copy=True, order='K', subok=False, ndmin=0, ndmax=0, like=None), and it returns an ndarray. The ones you will use most:

• dtype: if you do not give one, NumPy tries to use a default dtype that can represent the values, applying promotion rules when necessary. To force a type, for example complex: np.array([1, 2, 3], dtype=np.complex128) returns array([ 1.+0.j, 2.+0.j, 3.+0.j]).
• ndmin: the minimum number of dimensions, with ones prepended to the shape as needed. np.array([1, 2, 3], ndmin=2) returns array([[1, 2, 3]]). The default is 0.
• copy: the default is True. None copies only when needed, and False raises a ValueError when a copy cannot be avoided.
• subok: default False. ndmax is marked “New in 2.4.0” and like “New in 1.20.0” in the reference page, so older installs may not accept them.

If you pass a scalar instead of a sequence, you get a 0-dimensional array containing it, which matters again in the conversion section below.

Read the array: shape, ndim, size and dtype

Four attributes describe what you created. The quickstart defines them as follows, using np.arange(15).reshape(3, 5) as its example.

Attribute	Meaning (NumPy quickstart)	For np.arange(15).reshape(3, 5)
shape	Tuple giving the size of the array in each dimension	(3, 5)
ndim	Number of axes (dimensions)	2
size	Total number of elements, the product of the elements of shape	15
dtype	The type of the elements	an integer type

The quickstart shows the deduced type for a list of integers and a list of floats: np.array([2, 3, 4]).dtype is, for example, dtype('int64') and np.array([1.2, 3.5, 5.1]).dtype is dtype('float64'). The integer width can vary by platform, so do not hard-code it in tests.

Convert a NumPy array to a list (numpy array to list)

Call .tolist() on the array. According to the reference page, ndarray.tolist() takes no parameters and returns the array “as an a.ndim-levels deep nested list of Python scalars.” It returns a copy, and each item is converted to the nearest built-in Python type.

1. Create an array: a = np.array([[1, 2], [3, 4]]).
2. Convert it: a.tolist() returns [[1, 2], [3, 4]], a list of lists of Python int values.

list(arr) versus arr.tolist()

Both give you a Python list, but the contents differ. The tolist documentation compares them directly.

Case	list(a)	a.tolist()
1-D: a = np.uint32([1, 2])	[np.uint32(1), np.uint32(2)], items are of type numpy.uint32	[1, 2], items are Python int
2-D: [[1, 2], [3, 4]]	[array([1, 2]), array([3, 4])], a list of arrays	[[1, 2], [3, 4]], a list of lists
0-D: np.array(1)	raises TypeError: iteration over a 0-d array	returns 1, a plain Python scalar and not a list

The practical rule that follows: use tolist() when you need pure Python objects (for example before handing data to code that expects built-in types), and use list(a) only when you want to iterate over the first axis and are happy to receive NumPy scalars or sub-arrays.

Round trip: list, array, list

The reference page notes that “The array may be recreated via a = np.array(a.tolist()), although this may sometimes lose precision.” Treat the round trip as safe for display and hand-off, and compare values explicitly if exact precision matters.

Troubleshooting

TypeError: array() takes from 1 to 2 positional arguments

The quickstart lists this as a common error. np.array(1, 2, 3, 4) raises TypeError: array() takes from 1 to 2 positional arguments but 4 were given, because array expects one sequence, not several numbers. The fix is to wrap them in a list: np.array([1, 2, 3, 4]).

ValueError with copy=False

With copy=False, NumPy raises a ValueError if a copy cannot be avoided. Use copy=None to copy only when needed.

TypeError: iteration over a 0-d array

This appears when you call list() on a 0-dimensional array, such as np.array(1). Use .tolist(), which returns the scalar.

Unexpected dtype

If the dtype is not what you wanted, state it: np.array(data, dtype=...). Without it, NumPy chooses a type that can represent all values, so one float in an otherwise integer list makes the whole array float.

Related calls in NumPy and Python

The calls below are the ones used in this guide.

Task	Call	Returns
Build an array from a list or nested list	np.array(seq)	ndarray, dimensions follow the nesting
Build a numbered array and give it a shape	np.arange(15).reshape(3, 5)	ndarray with shape (3, 5)
Array to nested list of Python scalars	a.tolist()	list, a.ndim levels deep
Iterate the first axis into a list	list(a)	list of NumPy scalars (1-D) or sub-arrays (n-D)

For other array-building and array-selection calls, see NumPy linspace for evenly spaced values and NumPy where for conditional selection. NumPy itself is open source software.

Frequently asked questions

How do I make a NumPy array in Python?

Import NumPy as np and call np.array() with a single list, for example np.array([1, 2, 3]). Nested lists give a multidimensional array.

What is the difference between a list and a NumPy array?

Per the NumPy quickstart, a NumPy array is homogeneous: all elements share one type, and it is indexed by a tuple of non-negative integers. Its type is deduced from the elements you pass in.

How do I convert a NumPy array to a list?

Use a.tolist(). It returns a nested list of Python scalars, with one level of nesting per dimension.

Why does list(arr) give NumPy types but arr.tolist() gives Python types?

list(arr) iterates the array and keeps each element as a NumPy object, while tolist() converts every item to the nearest built-in Python type.

Does converting a list to an array and back lose data?

The NumPy documentation says recreating an array from a.tolist() may sometimes lose precision, so verify values when exactness matters.

Sources

• NumPy documentation, numpy.array (checked 2026-10-29)
• NumPy documentation, numpy.ndarray.tolist (checked 2026-10-29)
• NumPy documentation, NumPy quickstart (checked 2026-10-29)
• NumPy, Installing NumPy (checked 2026-10-29)