diff --git a/changelog.rst b/changelog.rst
index 81fbefe78..ce3602e4f 100644
--- a/changelog.rst
+++ b/changelog.rst
@@ -6,6 +6,14 @@ Features:
 * Add a ``\service`` prompt token for connections made using ``service=...``
   or ``PGSERVICE``. For example, set ``prompt = '\service> '``.
 
+Bug fixes:
+----------
+* Fix `-t`/`--tuples-only` hiding SQL errors. The error of a failed
+  statement was printed in place of the status footer that `-t` suppresses,
+  so a broken query printed nothing at all. Errors now go to stderr under
+  `-t`, as with `psql -t`, so stdout still carries only rows.
+
+
 4.7.1 (2026-09-20)
 ==================
 
diff --git a/pgcli/main.py b/pgcli/main.py
index 994f116e5..c38dfabbd 100644
--- a/pgcli/main.py
+++ b/pgcli/main.py
@@ -1386,6 +1386,14 @@ def _evaluate_command(self, text):
             ):
                 title = None
 
+            if not success and status and self.tuples_only:
+                # For a failed statement `status` is the error message, not
+                # the "SELECT 3" footer that -t drops, so -t used to hide the
+                # error completely. With -t stdout carries only rows, so the
+                # error goes to stderr, as psql does.
+                click.echo(status, err=True)
+                status = None
+
             execution = time() - start
             formatted = format_output(title, cur, headers, status, settings, self.explain_mode)
 
diff --git a/tests/features/command_option.feature b/tests/features/command_option.feature
index 38f5fef51..0f9403c92 100644
--- a/tests/features/command_option.feature
+++ b/tests/features/command_option.feature
@@ -22,6 +22,10 @@ Feature: run the cli with -c/--command option,
       then we see an error message
       and pgcli exits successfully
 
+  Scenario: run pgcli with -t and an invalid query
+     When we run pgcli with -t -c "SELECT invalid_column FROM nonexistent_table"
+      then we see the error on stderr and nothing on stdout
+
   Scenario: run pgcli with -c and multiple statements
      When we run pgcli with -c "SELECT 1; SELECT 2"
       then we see both query results
diff --git a/tests/features/steps/command_option.py b/tests/features/steps/command_option.py
index b783dc7a7..3ffe67e98 100644
--- a/tests/features/steps/command_option.py
+++ b/tests/features/steps/command_option.py
@@ -142,3 +142,34 @@ def step_see_all_command_outputs(context):
     # For the 3-command test, also check for third
     if "third" in output or "result" in output:
         assert "third" in output, f"Expected 'third' in output for 3-command test, but got: {output}"
+
+
+@when('we run pgcli with -t -c "{command}"')
+def step_run_pgcli_with_t_and_c(context, command):
+    """Run pgcli with -t and -c, keeping stdout and stderr apart."""
+    cmd = [
+        "pgcli",
+        "-h",
+        context.conf["host"],
+        "-p",
+        str(context.conf["port"]),
+        "-U",
+        context.conf["user"],
+        "-d",
+        context.conf["dbname"],
+        "-t",
+        "-c",
+        command,
+    ]
+    result = subprocess.run(cmd, cwd=context.package_root, capture_output=True, timeout=5)
+    context.cmd_output = result.stdout
+    context.cmd_stderr = result.stderr.decode()
+    context.exit_code = result.returncode
+
+
+@then("we see the error on stderr and nothing on stdout")
+def step_error_on_stderr_only(context):
+    """With -t, stdout carries only rows, so the error must be on stderr."""
+    stdout = context.cmd_output.decode()
+    assert "does not exist" in context.cmd_stderr, f"Expected the error on stderr, got: {context.cmd_stderr!r}"
+    assert "does not exist" not in stdout, f"Expected no error on stdout, got: {stdout!r}"
diff --git a/tests/test_main.py b/tests/test_main.py
index decae97b1..334221b1a 100644
--- a/tests/test_main.py
+++ b/tests/test_main.py
@@ -1087,3 +1087,23 @@ def test_connect_timeout_config_value_must_be_a_number(tmpdir):
         f.write("[main]\nconnect_timeout = soon\n")
     with pytest.raises(ValueError):
         PGCli(pgclirc_file=rc)
+
+
+@dbtest
+def test_tuples_only_sends_errors_to_stderr(executor, capsys):
+    # The error of a failed statement travels in `status`, the footer -t
+    # drops, so -t used to print nothing at all for a broken query.
+    cli = PGCli(pgexecute=executor, tuples_only=True)
+    output, query = cli._evaluate_command("select nope")
+    assert not query.successful
+    assert 'column "nope" does not exist' in capsys.readouterr().err
+    assert not any("nope" in line for line in output)
+
+
+@dbtest
+def test_tuples_only_rows_stay_on_stdout(executor, capsys):
+    cli = PGCli(pgexecute=executor, tuples_only=True)
+    output, query = cli._evaluate_command("select 42")
+    assert query.successful
+    assert any("42" in line for line in output)
+    assert capsys.readouterr().err == ""