diff --git a/pybigquery/sqlalchemy_bigquery.py b/pybigquery/sqlalchemy_bigquery.py
index 74573458..e074c82d 100644
--- a/pybigquery/sqlalchemy_bigquery.py
+++ b/pybigquery/sqlalchemy_bigquery.py
@@ -205,6 +205,9 @@ def visit_text(self, type_, **kw):
     def visit_string(self, type_, **kw):
         return 'STRING'
 
+    def visit_ARRAY(self, type_, **kw):
+        return "ARRAY<{}>".format(self.process(type_.item_type, **kw))
+
     def visit_BINARY(self, type_, **kw):
         return 'BYTES'
 
diff --git a/test/test_sqlalchemy_bigquery.py b/test/test_sqlalchemy_bigquery.py
index 831ee351..2ed998eb 100644
--- a/test/test_sqlalchemy_bigquery.py
+++ b/test/test_sqlalchemy_bigquery.py
@@ -358,6 +358,16 @@ def test_compiled_query_literal_binds(engine, engine_using_test_dataset, table,
     assert len(result) > 0
 
 
+@pytest.mark.parametrize(["column", "processed"], [
+    (types.String(), "STRING"),
+    (types.NUMERIC(), "NUMERIC"),
+    (types.ARRAY(types.String), "ARRAY<STRING>"),
+])
+def test_compile_types(engine, column, processed):
+    result = engine.dialect.type_compiler.process(column)
+    assert result == processed
+
+
 def test_joins(session, table, table_one_row):
     result = (session.query(table.c.string, func.count(table_one_row.c.integer))
                      .join(table_one_row, table_one_row.c.string == table.c.string)