diff --git a/lib/decimal.ex b/lib/decimal.ex
index b405079..004c205 100644
--- a/lib/decimal.ex
+++ b/lib/decimal.ex
@@ -2463,7 +2463,11 @@ defmodule Decimal do
 
   defp increment?(:down, _, _, _, _), do: false
 
-  defp increment?(:up, _, _, _, _), do: true
+  # Round-up is away from zero, but per the General Decimal Arithmetic spec
+  # an all-zero discarded part leaves the value unchanged (an exact value is
+  # never rounded up). Check the discarded digits like :ceiling/:floor do,
+  # rather than incrementing unconditionally.
+  defp increment?(:up, _, _, remain, sticky?), do: any_nonzero?(remain, sticky?)
 
   defp increment?(:ceiling, sign, _, remain, sticky?),
     do: sign == 1 and any_nonzero?(remain, sticky?)
diff --git a/test/decimal/context_test.exs b/test/decimal/context_test.exs
index 4e5a894..a42f42d 100644
--- a/test/decimal/context_test.exs
+++ b/test/decimal/context_test.exs
@@ -82,6 +82,17 @@ defmodule Decimal.ContextTest do
       assert Decimal.add(~d"0", ~d"-102") == d(-1, 11, 1)
       assert Decimal.add(~d"0", ~d"1.1") == d(1, 11, -1)
     end)
+
+    # :up rounds away from zero only when a discarded digit is nonzero. An
+    # exact value whose dropped digits are all zero must be left unchanged
+    # (matches the General Decimal Arithmetic spec and Python's decimal).
+    Context.with(%Context{precision: 1, rounding: :up}, fn ->
+      assert Decimal.add(~d"0", ~d"9.0") == d(1, 9, 0)
+      assert Decimal.add(~d"0", ~d"-9.0") == d(-1, 9, 0)
+      assert Decimal.mult(~d"9.00", ~d"1") == d(1, 9, 0)
+      # a nonzero discarded digit still rounds up
+      assert Decimal.add(~d"0", ~d"3.1") == d(1, 4, 0)
+    end)
   end
 
   test "with_context/2: large exponent gap addition" do