diff --git a/README.rst b/README.rst
index 491d50ca..e5d417c8 100644
--- a/README.rst
+++ b/README.rst
@@ -28,6 +28,7 @@ Here's a snippet from `Cram's own test suite`_::
         -y, --yes           answer yes to all questions
         -n, --no            answer no to all questions
         -E, --preserve-env  don't reset common environment variables
+	-e, --no-err-files  don't write .err files on test failures
         --keep-tmpdir       keep temporary directories
         --shell=PATH        shell to use for running tests (default: /bin/sh)
         --shell-opts=OPTS   arguments to invoke shell with
@@ -132,9 +133,10 @@ For example, if we run Cram on `its own example tests`_::
     s.
     # Ran 6 tests, 2 skipped, 1 failed.
 
-Cram will also write the test with its actual output to
-``examples/fail.t.err``, allowing you to use other diff tools. This
-file is automatically removed the next time the test passes.
+Unless run with ``-e`` or ``--no-err-files``, Cram will also write the
+test with its actual output to ``examples/fail.t.err``, allowing you to
+use other diff tools. This file is automatically removed the next time
+the test passes.
 
 When you're first writing a test, you might just write the commands
 and run the test to see what happens. If you run Cram with ``-i`` or
diff --git a/cram/_cli.py b/cram/_cli.py
index cad8381c..e9afe766 100644
--- a/cram/_cli.py
+++ b/cram/_cli.py
@@ -53,7 +53,8 @@ def _patch(cmd, diff):
     out, retcode = execute([cmd, '-p0'], stdin=b''.join(diff))
     return retcode == 0
 
-def runcli(tests, quiet=False, verbose=False, patchcmd=None, answer=None):
+def runcli(tests, quiet=False, verbose=False, patchcmd=None, answer=None,
+           noerrfiles=False):
     """Run tests with command line interface input/output.
 
     tests should be a sequence of 2-tuples containing the following:
@@ -101,12 +102,13 @@ def testwrapper():
                 if not quiet:
                     _log('\n', None, verbose)
 
-                errfile = open(errpath, 'wb')
-                try:
-                    for line in postout:
-                        errfile.write(line)
-                finally:
-                    errfile.close()
+                if not noerrfiles:
+                    errfile = open(errpath, 'wb')
+                    try:
+                        for line in postout:
+                            errfile.write(line)
+                    finally:
+                        errfile.close()
 
                 if not quiet:
                     origdiff = diff
@@ -119,7 +121,8 @@ def testwrapper():
                         _prompt('Accept this change?', 'yN', answer) == 'y'):
                         if _patch(patchcmd, diff):
                             _log(None, path + b': merged output\n', verbose)
-                            os.remove(errpath)
+                            if not noerrfiles:
+                                os.remove(errpath)
                         else:
                             _log(path + b': merge failed\n')
 
diff --git a/cram/_main.py b/cram/_main.py
index b6d80edc..d47d21b2 100644
--- a/cram/_main.py
+++ b/cram/_main.py
@@ -94,6 +94,8 @@ def _parseopts(args):
                  help='answer no to all questions')
     p.add_option('-E', '--preserve-env', action='store_true',
                  help="don't reset common environment variables")
+    p.add_option('-e', '--no-err-files', action='store_true',
+                 help="don't write .err files on test failures")
     p.add_option('--keep-tmpdir', action='store_true',
                  help='keep temporary directories')
     p.add_option('--shell', action='store', default='/bin/sh', metavar='PATH',
@@ -183,10 +185,12 @@ def main(args):
     os.mkdir(proctmp)
     try:
         tests = runtests(paths, tmpdirb, shell, indent=opts.indent,
-                         cleanenv=not opts.preserve_env, debug=opts.debug)
+                         cleanenv=not opts.preserve_env, debug=opts.debug,
+                         noerrfiles=opts.no_err_files)
         if not opts.debug:
             tests = runcli(tests, quiet=opts.quiet, verbose=opts.verbose,
-                           patchcmd=patchcmd, answer=answer)
+                           patchcmd=patchcmd, answer=answer,
+                           noerrfiles=opts.no_err_files)
             if opts.xunit_file is not None:
                 tests = runxunit(tests, opts.xunit_file)
 
diff --git a/cram/_run.py b/cram/_run.py
index 43813a45..4a609c7c 100644
--- a/cram/_run.py
+++ b/cram/_run.py
@@ -30,7 +30,8 @@ def _findtests(paths):
         else:
             yield os.path.normpath(p)
 
-def runtests(paths, tmpdir, shell, indent=2, cleanenv=True, debug=False):
+def runtests(paths, tmpdir, shell, indent=2, cleanenv=True, debug=False,
+             noerrfiles=False):
     """Run tests and yield results.
 
     This yields a sequence of 2-tuples containing the following:
@@ -69,7 +70,7 @@ def test():
                 os.chdir(testdir)
                 return testfile(abspath, shell, indent=indent,
                                 cleanenv=cleanenv, debug=debug,
-                                testname=path)
+                                testname=path, noerrfile=noerrfiles)
             finally:
                 os.chdir(cwd)
 
diff --git a/cram/_test.py b/cram/_test.py
index 2cec870d..3293f39c 100644
--- a/cram/_test.py
+++ b/cram/_test.py
@@ -21,7 +21,7 @@ def _escape(s):
             b' (esc)\n')
 
 def test(lines, shell='/bin/sh', indent=2, testname=None, env=None,
-         cleanenv=True, debug=False):
+         cleanenv=True, debug=False, noerrfile=False):
     r"""Run test lines and return input, output, and diff.
 
     This returns a 3-tuple containing the following:
@@ -177,7 +177,7 @@ def test(lines, shell='/bin/sh', indent=2, testname=None, env=None,
     return refout, postout, []
 
 def testfile(path, shell='/bin/sh', indent=2, env=None, cleanenv=True,
-             debug=False, testname=None):
+             debug=False, testname=None, noerrfile=False):
     """Run test at path and return input, output, and diff.
 
     This returns a 3-tuple containing the following:
@@ -218,6 +218,6 @@ def testfile(path, shell='/bin/sh', indent=2, env=None, cleanenv=True,
         if testname is None: # pragma: nocover
             testname = os.path.basename(abspath)
         return test(f, shell, indent=indent, testname=testname, env=env,
-                    cleanenv=cleanenv, debug=debug)
+                    cleanenv=cleanenv, debug=debug, noerrfile=noerrfile)
     finally:
         f.close()
diff --git a/tests/test.t b/tests/test.t
index e1f74b12..387ba490 100644
--- a/tests/test.t
+++ b/tests/test.t
@@ -151,3 +151,25 @@ Test running tests with the same filename in different directories:
   
   # Ran 2 tests, 0 skipped, 2 failed.
   [1]
+
+Test failing a test in a read-only directory with the --no-err-files option:
+
+  $ mkdir subdir
+  $ cat > subdir/test.t <<EOF
+  >   $ echo 1
+  > EOF
+  $ chmod a-w subdir
+  $ cram subdir >/dev/null 2>&1
+  [1]
+  $ cram --no-err-files subdir
+  !
+  --- subdir/test.t
+  +++ subdir/test.t.err
+  @@ -1,1 +1,2 @@
+     $ echo 1
+  +  1
+  
+  # Ran 1 tests, 0 skipped, 1 failed.
+  [1]
+
+  $ chmod a+w subdir
diff --git a/tests/usage.t b/tests/usage.t
index c193d7bb..44ed6fd0 100644
--- a/tests/usage.t
+++ b/tests/usage.t
@@ -17,6 +17,7 @@ Usage:
     -y, --yes           answer yes to all questions
     -n, --no            answer no to all questions
     -E, --preserve-env  don't reset common environment variables
+    -e, --no-err-files  don't write .err files on test failures
     --keep-tmpdir       keep temporary directories
     --shell=PATH        shell to use for running tests (default: /bin/sh)
     --shell-opts=OPTS   arguments to invoke shell with