diff --git a/sqlalchemy_firebird/firebird.py b/sqlalchemy_firebird/firebird.py
index f518509..4bdb134 100644
--- a/sqlalchemy_firebird/firebird.py
+++ b/sqlalchemy_firebird/firebird.py
@@ -10,6 +10,8 @@
     This driver uses new Firebird OO API provided by fbclient library.
 """  # noqa
 
+import sys
+
 from datetime import datetime
 from datetime import time
 from math import modf
@@ -140,8 +142,29 @@ def do_execute(self, cursor, statement, parameters, context=None):
         # Firebird-driver needs special time zone handling.
         #   https://github.com/FirebirdSQL/python3-driver/issues/19#issuecomment-1523045743
         adapted_parameters = [self.adapt_timezone(p) for p in parameters]
+        self._disable_blob_streaming(cursor)
         super().do_execute(cursor, statement, adapted_parameters, context)
 
+    def do_executemany(self, cursor, statement, parameters, context=None):
+        self._disable_blob_streaming(cursor)
+        super().do_executemany(cursor, statement, parameters, context)
+
+    def do_execute_no_params(self, cursor, statement, context=None):
+        self._disable_blob_streaming(cursor)
+        super().do_execute_no_params(cursor, statement, context)
+
+    @staticmethod
+    def _disable_blob_streaming(cursor):
+        # SQLAlchemy fully consumes a cursor and then closes it before its
+        # result processors run. firebird-driver closes any BlobReader
+        # objects together with the cursor, which would make BLOB columns
+        # unreadable from within SQLAlchemy result rows (issue #58). Force
+        # all BLOBs to be returned as fully materialized bytes/str by
+        # raising the per-cursor streaming threshold past any practical
+        # size; this leaves driver_config.stream_blob_threshold untouched.
+        if hasattr(cursor, "stream_blob_threshold"):
+            cursor.stream_blob_threshold = sys.maxsize
+
 
 def remove_keys(d, keys):
     return {x: d[x] for x in d if x not in keys}
diff --git a/test/test_issues.py b/test/test_issues.py
index 6f5a07a..14f5474 100644
--- a/test/test_issues.py
+++ b/test/test_issues.py
@@ -1,4 +1,4 @@
-from sqlalchemy import Column, select
+from sqlalchemy import Column, Integer, select
 from sqlalchemy import Table
 from sqlalchemy import testing
 from sqlalchemy.testing import eq_
@@ -9,6 +9,12 @@
 TEST_UNICODE = "próf-áêïôù-🗄️.fdb"
 TEST_BINARY = TEST_UNICODE.encode("utf-8")
 
+# firebird-driver returns BlobReader objects for BLOBs larger than
+# stream_blob_threshold (default 64 KiB). Use 200 KiB to trigger the
+# streaming path that issue #58 was about.
+LARGE_BINARY = b"\xa5" * (200 * 1024)
+LARGE_TEXT = ("loremipsum-" * 20_000)[: 200 * 1024]
+
 
 class IssuesTest(fixtures.TestBase):
     @testing.provide_metadata
@@ -36,3 +42,35 @@ def test_issue_76(self, connection, type_, expected):
             connection.execute(select(the_blob.c.the_value)).scalar(),
             expected,
         )
+
+    @testing.provide_metadata
+    @testing.combinations(
+        (fb_types.FBBLOB, LARGE_BINARY),
+        (fb_types.FBTEXT, LARGE_TEXT),
+        argnames="type_, expected",
+        id_="na",
+    )
+    def test_issue_58_large_blob(self, connection, type_, expected):
+        # Regression test for issue #58: BLOBs above firebird-driver's
+        # stream_blob_threshold were returned as BlobReader objects that
+        # SQLAlchemy could not consume (and which got closed alongside
+        # the cursor before the result processors ran).
+        metadata = self.metadata
+
+        large_blob = Table(
+            "large_blob",
+            metadata,
+            Column("id", Integer, primary_key=True),
+            Column("the_value", type_),
+        )
+        metadata.create_all(testing.db)
+
+        connection.execute(
+            large_blob.insert().values(id=1, the_value=expected)
+        )
+
+        result = connection.execute(select(large_blob.c.the_value)).scalar()
+
+        eq_(type(result), type(expected))
+        eq_(len(result), len(expected))
+        eq_(result, expected)