diff --git a/python/pyarrow/src/arrow/python/inference.cc b/python/pyarrow/src/arrow/python/inference.cc
index 291cc421495..ea1514be5be 100644
--- a/python/pyarrow/src/arrow/python/inference.cc
+++ b/python/pyarrow/src/arrow/python/inference.cc
@@ -127,6 +127,7 @@ class NumPyDtypeUnifier {
       _NUMPY_UNIFY_PROMOTE(INT16);
       _NUMPY_UNIFY_PROMOTE(INT32);
       _NUMPY_UNIFY_PROMOTE(INT64);
+      _NUMPY_UNIFY_PROMOTE(FLOAT16);
       _NUMPY_UNIFY_PROMOTE(FLOAT32);
       _NUMPY_UNIFY_PROMOTE(FLOAT64);
       default:
@@ -140,6 +141,7 @@ class NumPyDtypeUnifier {
       _NUMPY_UNIFY_PROMOTE(INT32);
       _NUMPY_UNIFY_PROMOTE(INT64);
       _NUMPY_UNIFY_NOOP(UINT8);
+      _NUMPY_UNIFY_PROMOTE_TO(FLOAT16, FLOAT32);
       _NUMPY_UNIFY_PROMOTE(FLOAT32);
       _NUMPY_UNIFY_PROMOTE(FLOAT64);
       default:
@@ -155,6 +157,7 @@ class NumPyDtypeUnifier {
       _NUMPY_UNIFY_PROMOTE(INT64);
       _NUMPY_UNIFY_NOOP(UINT8);
       _NUMPY_UNIFY_NOOP(UINT16);
+      _NUMPY_UNIFY_PROMOTE_TO(FLOAT16, FLOAT64);
       _NUMPY_UNIFY_PROMOTE_TO(FLOAT32, FLOAT64);
       _NUMPY_UNIFY_PROMOTE(FLOAT64);
       default:
@@ -171,6 +174,7 @@ class NumPyDtypeUnifier {
       _NUMPY_UNIFY_NOOP(UINT8);
       _NUMPY_UNIFY_NOOP(UINT16);
       _NUMPY_UNIFY_NOOP(UINT32);
+      _NUMPY_UNIFY_PROMOTE_TO(FLOAT16, FLOAT64);
       _NUMPY_UNIFY_PROMOTE_TO(FLOAT32, FLOAT64);
       _NUMPY_UNIFY_PROMOTE(FLOAT64);
       default:
@@ -183,6 +187,7 @@ class NumPyDtypeUnifier {
       _NUMPY_UNIFY_PROMOTE(UINT16);
       _NUMPY_UNIFY_PROMOTE(UINT32);
       _NUMPY_UNIFY_PROMOTE(UINT64);
+      _NUMPY_UNIFY_PROMOTE(FLOAT16);
       _NUMPY_UNIFY_PROMOTE(FLOAT32);
       _NUMPY_UNIFY_PROMOTE(FLOAT64);
       default:
@@ -195,6 +200,7 @@ class NumPyDtypeUnifier {
       _NUMPY_UNIFY_NOOP(UINT8);
       _NUMPY_UNIFY_PROMOTE(UINT32);
       _NUMPY_UNIFY_PROMOTE(UINT64);
+      _NUMPY_UNIFY_PROMOTE_TO(FLOAT16, FLOAT32);
       _NUMPY_UNIFY_PROMOTE(FLOAT32);
       _NUMPY_UNIFY_PROMOTE(FLOAT64);
       default:
@@ -207,6 +213,7 @@ class NumPyDtypeUnifier {
       _NUMPY_UNIFY_NOOP(UINT8);
       _NUMPY_UNIFY_NOOP(UINT16);
       _NUMPY_UNIFY_PROMOTE(UINT64);
+      _NUMPY_UNIFY_PROMOTE_TO(FLOAT16, FLOAT64);
       _NUMPY_UNIFY_PROMOTE_TO(FLOAT32, FLOAT64);
       _NUMPY_UNIFY_PROMOTE(FLOAT64);
       default:
@@ -219,6 +226,7 @@ class NumPyDtypeUnifier {
       _NUMPY_UNIFY_NOOP(UINT8);
       _NUMPY_UNIFY_NOOP(UINT16);
       _NUMPY_UNIFY_NOOP(UINT32);
+      _NUMPY_UNIFY_PROMOTE_TO(FLOAT16, FLOAT64);
       _NUMPY_UNIFY_PROMOTE_TO(FLOAT32, FLOAT64);
       _NUMPY_UNIFY_PROMOTE(FLOAT64);
       default:
@@ -228,6 +236,14 @@ class NumPyDtypeUnifier {
 
   int Observe_FLOAT16(PyArray_Descr* descr, int dtype) {
     switch (dtype) {
+      _NUMPY_UNIFY_NOOP(INT8);
+      _NUMPY_UNIFY_NOOP(UINT8);
+      _NUMPY_UNIFY_PROMOTE_TO(INT16, FLOAT32);
+      _NUMPY_UNIFY_PROMOTE_TO(UINT16, FLOAT32);
+      _NUMPY_UNIFY_PROMOTE_TO(INT32, FLOAT64);
+      _NUMPY_UNIFY_PROMOTE_TO(UINT32, FLOAT64);
+      _NUMPY_UNIFY_PROMOTE_TO(INT64, FLOAT64);
+      _NUMPY_UNIFY_PROMOTE_TO(UINT64, FLOAT64);
       _NUMPY_UNIFY_PROMOTE(FLOAT32);
       _NUMPY_UNIFY_PROMOTE(FLOAT64);
       default:
diff --git a/python/pyarrow/tests/test_convert_builtin.py b/python/pyarrow/tests/test_convert_builtin.py
index 3ac7be88eaa..9ecdcda51f7 100644
--- a/python/pyarrow/tests/test_convert_builtin.py
+++ b/python/pyarrow/tests/test_convert_builtin.py
@@ -481,6 +481,20 @@ def test_numpy_scalars_mixed_type():
     assert arr.equals(expected)
 
 
+@pytest.mark.numpy
+@pytest.mark.parametrize("int_type", [
+    "int8", "int16", "int32", "int64", "uint8", "uint16", "uint32", "uint64"
+])
+def test_numpy_scalars_mixed_float16_int(int_type):
+    # GH-25859: follow NumPy type promotion
+    int_scalar = getattr(np, int_type)(5)
+    expected_type = pa.from_numpy_dtype(np.result_type(np.float16, int_type))
+    for data in ([np.float16(1.5), int_scalar], [int_scalar, np.float16(1.5)]):
+        arr = pa.array(data)
+        assert arr.type == expected_type
+        assert arr.to_pylist() == [float(x) for x in data]
+
+
 @pytest.mark.numpy
 @pytest.mark.xfail(reason="Type inference for uint64 not implemented",
                    raises=OverflowError)