diff --git a/sklearn/cluster/_mean_shift.py b/sklearn/cluster/_mean_shift.py
index cbed1c5be3b9e..46a00ed3f0740 100644
--- a/sklearn/cluster/_mean_shift.py
+++ b/sklearn/cluster/_mean_shift.py
@@ -119,6 +119,7 @@ def _mean_shift_single_seed(my_mean, X, nbrs, max_iter):
     return tuple(my_mean), len(points_within), completed_iterations
 
 
+@validate_params({"X": ["array-like"]})
 def mean_shift(
     X,
     *,
@@ -141,9 +142,9 @@ def mean_shift(
         Input data.
 
     bandwidth : float, default=None
-        Kernel bandwidth.
+        Kernel bandwidth. If not None, must be in the range [0, +inf).
 
-        If bandwidth is not given, it is determined using a heuristic based on
+        If None, the bandwidth is determined using a heuristic based on
         the median of all pairwise distances. This will take quadratic time in
         the number of samples. The sklearn.cluster.estimate_bandwidth function
         can be used to do this more efficiently.
diff --git a/sklearn/tests/test_public_functions.py b/sklearn/tests/test_public_functions.py
index 9b2b56cdb3eb8..cc88bc0afcb48 100644
--- a/sklearn/tests/test_public_functions.py
+++ b/sklearn/tests/test_public_functions.py
@@ -155,6 +155,7 @@ def test_function_param_validation(func_module):
 
 PARAM_VALIDATION_CLASS_WRAPPER_LIST = [
     ("sklearn.cluster.affinity_propagation", "sklearn.cluster.AffinityPropagation"),
+    ("sklearn.cluster.mean_shift", "sklearn.cluster.MeanShift"),
     ("sklearn.cluster.spectral_clustering", "sklearn.cluster.SpectralClustering"),
     ("sklearn.covariance.ledoit_wolf", "sklearn.covariance.LedoitWolf"),
     ("sklearn.covariance.oas", "sklearn.covariance.OAS"),