diff --git a/src/narwhals/_arrow/namespace.py b/src/narwhals/_arrow/namespace.py
index 4641e91354..ec6260aae6 100644
--- a/src/narwhals/_arrow/namespace.py
+++ b/src/narwhals/_arrow/namespace.py
@@ -164,7 +164,11 @@ def func(df: ArrowDataFrame) -> list[ArrowSeries]:
             expr_results = tuple(chain.from_iterable(expr(df) for expr in exprs))
             series = [s.fill_null(0, strategy=None, limit=None) for s in expr_results]
             non_na = [1 - s.is_null().cast(int_64) for s in expr_results]
-            return [reduce(operator.add, series) / reduce(operator.add, non_na)]
+            count = reduce(operator.add, non_na)
+            mean = reduce(operator.add, series) / count
+            # A row where every input is null has a count of 0: return null, not 0/0 = NaN.
+            native = pc.if_else((count == 0).native, None, mean.native)
+            return [mean._with_native(native, preserve_broadcast=True)]
 
         return self._expr._from_callable(
             func=func,
diff --git a/src/narwhals/_duckdb/namespace.py b/src/narwhals/_duckdb/namespace.py
index 3541c6ef6b..2614d5fb11 100644
--- a/src/narwhals/_duckdb/namespace.py
+++ b/src/narwhals/_duckdb/namespace.py
@@ -153,7 +153,7 @@ def func(cols: Iterable[Expression]) -> Expression:
             count = reduce(
                 operator.add, (col.isnotnull().cast(duckdb_dtypes.BIGINT) for col in cols)
             )
-            return total / count
+            return when(count > lit(0), total / count)
 
         return self._expr._from_elementwise_horizontal_op(func, *exprs)
 
diff --git a/src/narwhals/_ibis/namespace.py b/src/narwhals/_ibis/namespace.py
index 595dc275f0..2970e0f1ed 100644
--- a/src/narwhals/_ibis/namespace.py
+++ b/src/narwhals/_ibis/namespace.py
@@ -136,7 +136,7 @@ def func(cols: Iterable[ir.Value]) -> ir.Value:
                     for col in cols
                 ),
             )
-            return numerator / denominator
+            return numerator / denominator.nullif(lit(0))
 
         return self._expr._from_elementwise_horizontal_op(func, *exprs)
 
diff --git a/src/narwhals/_spark_like/namespace.py b/src/narwhals/_spark_like/namespace.py
index 50d9d755bd..79cc442910 100644
--- a/src/narwhals/_spark_like/namespace.py
+++ b/src/narwhals/_spark_like/namespace.py
@@ -215,7 +215,7 @@ def func(cols: Iterable[Column]) -> Column:
                 operator.add,
                 (col.isNotNull().cast(self._native_dtypes.IntegerType()) for col in cols),
             )
-            return true_divide(F, numerator, denominator)
+            return F.when(denominator > F.lit(0), true_divide(F, numerator, denominator))
 
         return self._expr._from_elementwise_horizontal_op(func, *exprs)
 
diff --git a/tests/expr_and_series/mean_horizontal_test.py b/tests/expr_and_series/mean_horizontal_test.py
index a3f03ade2b..8e232517e1 100644
--- a/tests/expr_and_series/mean_horizontal_test.py
+++ b/tests/expr_and_series/mean_horizontal_test.py
@@ -29,3 +29,13 @@ def test_meanh_all(constructor: Constructor) -> None:
     result = df.select(c=nw.mean_horizontal(nw.all()))
     expected = {"c": [6, 12, 18]}
     assert_equal_data(result, expected)
+
+
+def test_meanh_all_null_row_is_null(constructor: Constructor) -> None:
+    data = {"a": [1.0, None], "b": [3.0, None]}
+    df = nw.from_native(constructor(data))
+    result = df.select(mean=nw.mean_horizontal("a", "b")).select(
+        nw.col("mean"), is_null=nw.col("mean").is_null()
+    )
+    expected = {"mean": [2.0, None], "is_null": [False, True]}
+    assert_equal_data(result, expected)