diff --git a/albumentations/augmentations/functional.py b/albumentations/augmentations/functional.py
index 9a221f599..ce30b9676 100644
--- a/albumentations/augmentations/functional.py
+++ b/albumentations/augmentations/functional.py
@@ -1512,10 +1512,8 @@ def fancy_pca(img, alpha=0.1):
     return orig_img
 
 
-@clipped
 def glass_blur(img, sigma, max_delta, iterations, dxy, mode):
-    coef = MAX_VALUES_BY_DTYPE[img.dtype]
-    x = np.uint8(cv2.GaussianBlur(np.array(img) / coef, sigmaX=sigma, ksize=(0, 0)) * coef)
+    x = cv2.GaussianBlur(np.array(img), sigmaX=sigma, ksize=(0, 0))
 
     if mode == "fast":
 
@@ -1542,7 +1540,7 @@ def glass_blur(img, sigma, max_delta, iterations, dxy, mode):
             dx = dxy[ind, i, 1]
             x[h, w], x[h + dy, w + dx] = x[h + dy, w + dx], x[h, w]
 
-    return np.clip(cv2.GaussianBlur(x / coef, sigmaX=sigma, ksize=(0, 0)), 0, 1) * coef
+    return cv2.GaussianBlur(x, sigmaX=sigma, ksize=(0, 0))
 
 
 def _adjust_brightness_torchvision_uint8(img, factor):
diff --git a/tests/test_transforms.py b/tests/test_transforms.py
index 4dcc4b9f9..6c26bf1c7 100644
--- a/tests/test_transforms.py
+++ b/tests/test_transforms.py
@@ -827,3 +827,27 @@ def test_shift_scale_separate_shift_x_shift_y(image, mask):
     )
     assert np.array_equal(data["image"], expected_image)
     assert np.array_equal(data["mask"], expected_mask)
+
+
+@pytest.mark.parametrize(["val_uint8"], [[0], [1], [128], [255]])
+def test_glass_blur_float_uint8_diff_less_than_two(val_uint8):
+
+    x_uint8 = np.zeros((5, 5)).astype(np.uint8)
+    x_uint8[2, 2] = val_uint8
+
+    x_float32 = np.zeros((5, 5)).astype(np.float32)
+    x_float32[2, 2] = val_uint8 / 255.0
+
+    glassblur = A.GlassBlur(always_apply=True, max_delta=1)
+
+    np.random.seed(0)
+    blur_uint8 = glassblur(image=x_uint8)["image"]
+
+    np.random.seed(0)
+    blur_float32 = glassblur(image=x_float32)["image"]
+
+    # Before comparison, rescale the blur_float32 to [0, 255]
+    diff = np.abs(blur_uint8 - blur_float32 * 255)
+
+    # The difference between the results of float32 and uint8 will be at most 2.
+    assert np.all(diff <= 2.0)