diff --git a/doc/whats_new/v1.1.rst b/doc/whats_new/v1.1.rst
index f0d00501d3368..57bb1bd36720b 100644
--- a/doc/whats_new/v1.1.rst
+++ b/doc/whats_new/v1.1.rst
@@ -173,6 +173,13 @@ Changelog
   instead of `__init__`.
   :pr:`21434` by :user:`Krum Arnaudov <krumeto>`.
 
+
+:mod:`sklearn.decomposition.KernelPCA`
+......................................
+- |Fix| :class:`decomposition.KernelPCA` now validates input parameters in
+  `fit` instead of `__init__`.
+  :pr:`21567` by :user:`Maggie Chege <MaggieChege>`.
+
 :mod:`sklearn.svm`
 ..................
 
diff --git a/sklearn/decomposition/_kernel_pca.py b/sklearn/decomposition/_kernel_pca.py
index 0efcf2d4fd341..bf08c7e215ccc 100644
--- a/sklearn/decomposition/_kernel_pca.py
+++ b/sklearn/decomposition/_kernel_pca.py
@@ -257,8 +257,6 @@ def __init__(
         copy_X=True,
         n_jobs=None,
     ):
-        if fit_inverse_transform and kernel == "precomputed":
-            raise ValueError("Cannot fit_inverse_transform with a precomputed kernel.")
         self.n_components = n_components
         self.kernel = kernel
         self.kernel_params = kernel_params
@@ -429,6 +427,8 @@ def fit(self, X, y=None):
         self : object
             Returns the instance itself.
         """
+        if self.fit_inverse_transform and self.kernel == "precomputed":
+            raise ValueError("Cannot fit_inverse_transform with a precomputed kernel.")
         X = self._validate_data(X, accept_sparse="csr", copy=self.copy_X)
         self._centerer = KernelCenterer()
         K = self._get_kernel(X)
diff --git a/sklearn/decomposition/tests/test_kernel_pca.py b/sklearn/decomposition/tests/test_kernel_pca.py
index 72b40ec83e308..848b9bb1b9bbf 100644
--- a/sklearn/decomposition/tests/test_kernel_pca.py
+++ b/sklearn/decomposition/tests/test_kernel_pca.py
@@ -78,8 +78,12 @@ def test_kernel_pca_invalid_parameters():
     Tests fitting inverse transform with a precomputed kernel raises a
     ValueError.
     """
-    with pytest.raises(ValueError):
-        KernelPCA(10, fit_inverse_transform=True, kernel="precomputed")
+    estimator = KernelPCA(
+        n_components=10, fit_inverse_transform=True, kernel="precomputed"
+    )
+    err_ms = "Cannot fit_inverse_transform with a precomputed kernel"
+    with pytest.raises(ValueError, match=err_ms):
+        estimator.fit(np.random.randn(10, 10))
 
 
 def test_kernel_pca_consistent_transform():
diff --git a/sklearn/tests/test_common.py b/sklearn/tests/test_common.py
index a777e0e716eb1..483c0c7641a63 100644
--- a/sklearn/tests/test_common.py
+++ b/sklearn/tests/test_common.py
@@ -410,7 +410,6 @@ def test_transformers_get_feature_names_out(transformer):
     "FeatureUnion",
     "GridSearchCV",
     "HalvingGridSearchCV",
-    "KernelPCA",
     "Pipeline",
     "SGDOneClassSVM",
     "TheilSenRegressor",