diff --git a/script/parse_lava_job.py b/script/parse_lava_job.py
new file mode 100755
index 0000000..9e331e3
--- /dev/null
+++ b/script/parse_lava_job.py
@@ -0,0 +1,104 @@
+#!/usr/bin/env python3
+#
+# Copyright (c) 2019, Arm Limited. All rights reserved.
+#
+# SPDX-License-Identifier: BSD-3-Clause
+#
+
+# After lava job is dispatched, its results will be collected in
+# $WORKSPACE/job_results.yaml file. Parse that file, and exit from this script
+# with the respective exit status
+
+import argparse
+import os
+import sys
+import yaml
+
+
+def report_job_failure():
+    job_url = os.environ["JOB_URL"]
+    build_number = os.environ["BUILD_NUMBER"]
+    print()
+    print("Job failed!")
+    print("See " + "/".join([job_url.rstrip("/"), build_number, "artifact",
+                             "job_output.log"]))
+    print()
+    sys.exit(1)
+
+
+def report_job_success():
+    print()
+    print("Job success.")
+    print()
+    sys.exit(0)
+
+
+def parse_cmd_line():
+    parser = argparse.ArgumentParser(description="Parse results from LAVA. "
+        "The results must be provided as a YAML file.")
+    parser.add_argument("--payload-type", default="linux", type=str,
+        help="Type of payload that was used in the test (default: %(default)s)")
+    parser.add_argument("--file",
+        default=os.path.join(os.environ["WORKSPACE"], "job_results.yaml"),
+        type=str, help="YAML file to parse (default: %(default)s)")
+    args = parser.parse_args()
+    return args
+
+
+args = parse_cmd_line()
+
+with open(args.file) as fd:
+    results = yaml.load(fd)
+
+    # Iterate through results. Find the element whose name is "job" in the
+    # "lava" suite. It contains the result of the overall LAVA run.
+    for phase in results:
+        if phase["name"] == "job" and phase["suite"] == "lava":
+            break
+    else:
+        raise Exception("Couldn't find 'job' phase in 'lava' suite in results")
+
+    if phase["result"] != "pass":
+        report_job_failure()
+
+    # If we've simply booted to the Linux shell prompt then we don't need to
+    # further analyze the results from LAVA.
+    if args.payload_type == "linux":
+        report_job_success()
+
+    # If we've run TFTF tests instead, then do some further parsing.
+    # First make sure the test session finished.
+    for phase in filter(lambda p: p["name"] == "lava-test-monitor", results):
+        if phase["result"] != "pass":
+            print("TFTF test session failed. Did it time out?")
+            report_job_failure()
+        break
+    else:
+        raise Exception("Couldn't find 'lava-test-monitor' phase results")
+
+    # Then count the number of tests that failed/skipped.
+    test_failures = 0
+    test_skips = 0
+    for phase in filter(lambda p: p["suite"] == "tftf", results):
+        metadata = phase["metadata"]
+        testcase_name = metadata["case"]
+        testcase_result = metadata["result"]
+        if testcase_result == "fail":
+            test_failures += 1
+            print("=> FAILED: " + testcase_name)
+        elif testcase_result == "skip":
+            test_skips += 1
+            print("   SKIPPED: " + testcase_name)
+
+    # Print a test summary
+    print()
+    if test_failures == 0 and test_skips == 0:
+        print("All tests passed.")
+    else:
+        print("{} tests failed; {} skipped. All other tests passed.".format(
+            test_failures, test_skips))
+
+    if test_failures == 0:
+        report_job_success()
+    else:
+        report_job_failure()