diff --git a/infer/src/backend/infer.ml b/infer/src/backend/infer.ml
index 8d31a4395..12ec50209 100644
--- a/infer/src/backend/infer.ml
+++ b/infer/src/backend/infer.ml
@@ -111,33 +111,37 @@ let () =
   let analysis_is_crashcontext = match Config.analyzer with
     | Some Crashcontext -> true
     | _ -> false in
-  if analysis_is_crashcontext then
-    (* check whether this is the top-level infer process *)
-    let top_level_infer =
-      (* if the '--buck' option was passed, then this is the top level process iff the build command
-         starts with 'buck' *)
-      if Config.buck then buck
-      (* otherwise, we assume javac as the build command and thus only one process *)
-      else true in
-    if top_level_infer then
-      (* if we are the top-level process, then find the output directory and collect all
-         crashcontext summaries under it in a single crashcontext.json file *)
-      let root_out_dir = if buck then begin
-          let project_root = match Config.project_root with
-            | Some root -> root
-            | None -> Filename.dirname Config.results_dir in
-          let buck_out = match Config.buck_out with
-            | Some dir -> dir
-            | None -> "buck-out" in
-          Filename.concat project_root buck_out
-        end
-        else Config.results_dir in
-      match Config.stacktrace with
-      | None -> failwith "Detected -a crashcontext without --stacktrace, \
-                          this should have been checked earlier."
-      | Some s -> Crashcontext.collect_all_summaries root_out_dir s;
-  if status <> Unix.WEXITED 0 then (
-    prerr_endline ("Failed to execute: " ^ (String.concat " " (Array.to_list args_py))) ;
-    exit 1
+  (if analysis_is_crashcontext then
+     (* check whether this is the top-level infer process *)
+     let top_level_infer =
+       (* if the '--buck' option was passed, then this is the top level process iff the build
+          command starts with 'buck' *)
+       if Config.buck then buck
+       (* otherwise, we assume javac as the build command and thus only one process *)
+       else true in
+     if top_level_infer then
+       (* if we are the top-level process, then find the output directory and collect all
+          crashcontext summaries under it in a single crashcontext.json file *)
+       let root_out_dir = if buck then begin
+           let project_root = match Config.project_root with
+             | Some root -> root
+             | None -> Filename.dirname Config.results_dir in
+           let buck_out = match Config.buck_out with
+             | Some dir -> dir
+             | None -> "buck-out" in
+           Filename.concat project_root buck_out
+         end
+         else Config.results_dir in
+       match Config.stacktrace with
+       | None -> failwith "Detected -a crashcontext without --stacktrace, \
+                           this should have been checked earlier."
+       | Some s -> Crashcontext.collect_all_summaries root_out_dir s
   );
-  ()
+  let exit_code = match status with
+    | Unix.WEXITED i -> i
+    | _ -> 1 in
+  if exit_code <> 0 then (
+    if not (exit_code = 2 && Config.fail_on_bug) then
+      prerr_endline ("Failed to execute: " ^ (String.concat " " (Array.to_list args_py))) ;
+    exit exit_code
+  )
diff --git a/infer/tests/build_systems/build_integration_tests.py b/infer/tests/build_systems/build_integration_tests.py
index db0daf0c0..f6ca49ac8 100755
--- a/infer/tests/build_systems/build_integration_tests.py
+++ b/infer/tests/build_systems/build_integration_tests.py
@@ -62,6 +62,8 @@ ALL_TESTS = [
     'buck',
     'cc1',
     'cmake',
+    'componentkit',
+    'fail',
     'gradle',
     'javac',
     'locale',
@@ -114,7 +116,7 @@ def save_report(reports, filename):
                             separators=(',', ': '), sort_keys=True)
 
 
-def run_analysis(clean_cmds, build_cmds, extra_check, env=None):
+def run_analysis(clean_cmds, build_cmds, extra_check, should_fail, env=None):
     for clean_cmd in clean_cmds:
         subprocess.check_call(clean_cmd, env=env)
 
@@ -134,7 +136,16 @@ def run_analysis(clean_cmds, build_cmds, extra_check, env=None):
                 mode='w',
                 suffix='.out',
                 prefix='analysis_') as analysis_output:
-            subprocess.check_call(infer_cmd, stdout=analysis_output, env=env)
+            try:
+                subprocess.check_call(infer_cmd,
+                                      stdout=analysis_output, env=env)
+                if should_fail is not None:
+                    # hacky since we should clean up infer-out, etc. as below
+                    # if you made the test fails, this is your punishment
+                    assert False
+            except subprocess.CalledProcessError, exn:
+                if exn.returncode != should_fail:
+                    raise
 
     json_path = os.path.join(temp_out_dir, REPORT_JSON)
     found_errors = utils.load_json_from_path(json_path)
@@ -228,6 +239,7 @@ def test(name,
          enabled=None,
          report_fname=None,
          extra_check=lambda x: None,
+         should_fail=None,
          preprocess=lambda: None,
          postprocess=lambda errors: errors):
     """Run a test.
@@ -244,6 +256,10 @@ def test(name,
     - [enabled] whether the test should attempt to run. By default it
       is enabled if [[name] in [to_test]]
     - [report_fname] where to find the expected Infer results
+    - [extra_check] some function that will be given the temporary
+      results directory as argument
+    - [should_fail] if not None then running infer is expected to fail
+      with [should_fail] error code
     - [preprocess] a function to run before the clean and compile
       commands. If the function returns something non-None, use that as
       the compile commands.
@@ -251,6 +267,7 @@ def test(name,
       modify them. It must return an Infer report.
 
     Returns [True] if the test ran, [False] otherwise.
+
     """
     # python can't into using values of arguments in the default
     # values of other arguments
@@ -282,6 +299,7 @@ def test(name,
         clean_commands,
         compile_commands,
         extra_check=extra_check,
+        should_fail=should_fail,
         env=env)
     original = os.path.join(EXPECTED_OUTPUTS_DIR, report_fname)
     do_test(postprocess(errors), original)
@@ -481,6 +499,13 @@ class BuildIntegrationTest(unittest.TestCase):
                            'TestIgnoreImports.mm'],
                'infer_args': ['--cxx', '--no-filtering']}])
 
+    def test_fail_on_issue(self):
+        test('fail', '--fail-on-issue flag',
+             CODETOANALYZE_DIR,
+             [{'compile': ['clang', '-c', 'hello.c'],
+               'infer_args': ['--fail-on-issue']}],
+             should_fail=2)
+
     def test_pmd_xml_output(self):
         def pmd_check(infer_out):
             assert os.path.exists(os.path.join(infer_out, 'report.xml'))
diff --git a/infer/tests/build_systems/expected_outputs/fail_report.json b/infer/tests/build_systems/expected_outputs/fail_report.json
new file mode 100644
index 000000000..a6ed10eff
--- /dev/null
+++ b/infer/tests/build_systems/expected_outputs/fail_report.json
@@ -0,0 +1,7 @@
+[
+  {
+    "bug_type": "NULL_DEREFERENCE",
+    "file": "hello.c",
+    "procedure": "test"
+  }
+]
\ No newline at end of file