diff --git a/cpp/src/arrow/compute/kernels/scalar_cast_numeric.cc b/cpp/src/arrow/compute/kernels/scalar_cast_numeric.cc
index ba2a8c4b5638..66d4d5985cad 100644
--- a/cpp/src/arrow/compute/kernels/scalar_cast_numeric.cc
+++ b/cpp/src/arrow/compute/kernels/scalar_cast_numeric.cc
@@ -486,8 +486,12 @@ struct DecimalConversions<Decimal256, InDecimal> {
 
 template <typename InDecimal>
 struct DecimalConversions<Decimal32, InDecimal> {
-  static Decimal32 ConvertInput(InDecimal&& val) { return Decimal32(val.low_bits()); }
-  static Decimal32 ConvertOutput(Decimal32&& val) { return val; }
+  // Scale then truncate. Narrowing in ConvertInput would drop coefficient bits
+  // before Rescale and FitsInPrecision could see them.
+  static InDecimal ConvertInput(InDecimal&& val) { return val; }
+  static Decimal32 ConvertOutput(InDecimal&& val) {
+    return Decimal32(static_cast<int32_t>(val.low_bits()));
+  }
 };
 
 template <>
diff --git a/cpp/src/arrow/compute/kernels/scalar_cast_test.cc b/cpp/src/arrow/compute/kernels/scalar_cast_test.cc
index 864ec4afce0c..56999295d792 100644
--- a/cpp/src/arrow/compute/kernels/scalar_cast_test.cc
+++ b/cpp/src/arrow/compute/kernels/scalar_cast_test.cc
@@ -1726,6 +1726,45 @@ TEST(Cast, Decimal128ToDecimal32) {
   }
 }
 
+// GH-53288: narrowing to Decimal32 must rescale and range-check the full-width
+// coefficient. Truncating to 32 bits first silently wrapped the value. One
+// specialization serves every source width, so all of them were affected.
+TEST(Cast, DecimalToDecimal32NarrowingChecksFullCoefficient) {
+  CastOptions options;
+  options.allow_decimal_truncate = false;
+
+  // Ten-digit coefficient does not fit decimal32(9, 0). Truncating first yielded
+  // 5000000000 - 2^32 = 705032704, which does fit, so the cast wrongly succeeded.
+  auto too_wide = ArrayFromJSON(decimal128(12, 0), R"(["5000000000."])");
+  options.to_type = decimal32(9, 0);
+  CheckCastFails(too_wide, options);
+
+  // 536870912000 is exactly 125 * 2^32, so the low 32 bits are zero. Rescaling by
+  // -3 first gives 536870912, which fits decimal32(9, -3) and must be preserved.
+  Decimal32Builder builder(decimal32(9, -3));
+  ASSERT_OK(builder.Append(Decimal32(536870912)));  // 536870912 * 10^3
+  ASSERT_OK_AND_ASSIGN(auto expected_negative_scale, builder.Finish());
+
+  auto negative_scale_input = ArrayFromJSON(decimal128(12, 0), R"(["536870912000."])");
+  CheckCast(negative_scale_input, expected_negative_scale, options);
+
+  // 30000000.00 has an unscaled value of 3000000000, which wraps to a negative
+  // int32. The wrapped value is no longer divisible by 100, so the exact rescale
+  // was rejected as a data loss.
+  auto exact_rescale = ArrayFromJSON(decimal128(12, 2), R"(["30000000.00"])");
+  CheckCast(exact_rescale, ArrayFromJSON(decimal32(9, 0), R"(["30000000."])"), options);
+
+  // The same specialization serves every source width, so check the other two.
+  options.to_type = decimal32(9, 0);
+  CheckCastFails(ArrayFromJSON(decimal64(18, 0), R"(["5000000000."])"), options);
+  CheckCastFails(ArrayFromJSON(decimal256(40, 0), R"(["5000000000."])"), options);
+
+  // With truncation allowed the rescale now also happens at the source width, so
+  // the downscale sees 536870912000 rather than its low 32 bits.
+  options.allow_decimal_truncate = true;
+  CheckCast(negative_scale_input, expected_negative_scale, options);
+}
+
 TEST(Cast, Decimal128ToDecimal64) {
   CastOptions options;