diff --git a/src/zarr/core/dtype/npy/float.py b/src/zarr/core/dtype/npy/float.py
index 0be2cbca9b..2a23cb429d 100644
--- a/src/zarr/core/dtype/npy/float.py
+++ b/src/zarr/core/dtype/npy/float.py
@@ -201,6 +201,16 @@ def _check_scalar(self, data: object) -> TypeGuard[FloatLike]:
         TypeGuard[FloatLike]
             True if the input is a valid scalar value, False otherwise.
         """
+        if isinstance(data, str):
+            # Only accept strings that are valid float representations (e.g. "NaN", "inf").
+            # Plain strings that cannot be converted should return False so that cast_scalar
+            # raises TypeError rather than a confusing ValueError.
+            try:
+                self.to_native_dtype().type(data)
+            except (ValueError, OverflowError):
+                return False
+            else:
+                return True
         return isinstance(data, FloatLike)
 
     def _cast_scalar_unchecked(self, data: FloatLike) -> TFloatScalar_co:
diff --git a/tests/test_dtype/test_npy/test_float.py b/tests/test_dtype/test_npy/test_float.py
index 1bbcbbc81f..8d8e768263 100644
--- a/tests/test_dtype/test_npy/test_float.py
+++ b/tests/test_dtype/test_npy/test_float.py
@@ -65,7 +65,10 @@ class TestFloat16(_BaseTestFloat):
         (Float16(), -1.0, np.float16(-1.0)),
         (Float16(), "NaN", np.float16("NaN")),
     )
-    invalid_scalar_params = ((Float16(), {"set!"}),)
+    invalid_scalar_params = (
+        (Float16(), {"set!"}),
+        (Float16(), "not_a_float"),
+    )
     hex_string_params = (("0x7fc0", np.nan), ("0x7fc1", np.nan), ("0x3c00", 1.0))
     item_size_params = (Float16(),)
 
@@ -113,7 +116,10 @@ class TestFloat32(_BaseTestFloat):
         (Float32(), -1.0, np.float32(-1.0)),
         (Float32(), "NaN", np.float32("NaN")),
     )
-    invalid_scalar_params = ((Float32(), {"set!"}),)
+    invalid_scalar_params = (
+        (Float32(), {"set!"}),
+        (Float32(), "not_a_float"),
+    )
     hex_string_params = (("0x7fc00000", np.nan), ("0x7fc00001", np.nan), ("0x3f800000", 1.0))
     item_size_params = (Float32(),)
 
@@ -160,7 +166,10 @@ class TestFloat64(_BaseTestFloat):
         (Float64(), -1.0, np.float64(-1.0)),
         (Float64(), "NaN", np.float64("NaN")),
     )
-    invalid_scalar_params = ((Float64(), {"set!"}),)
+    invalid_scalar_params = (
+        (Float64(), {"set!"}),
+        (Float64(), "not_a_float"),
+    )
     hex_string_params = (
         ("0x7ff8000000000000", np.nan),
         ("0x7ff8000000000001", np.nan),