diff --git a/base/interactiveutil.jl b/base/interactiveutil.jl
index 7f6a4854f5d3e..958b88a6ae464 100644
--- a/base/interactiveutil.jl
+++ b/base/interactiveutil.jl
@@ -697,15 +697,20 @@ end
 # testing
 
 """
-    runtests([tests=["all"] [, numcores=ceil(Int, Sys.CPU_CORES / 2) ]])
+    Base.runtests(tests=["all"], numcores=ceil(Int, Sys.CPU_CORES / 2);
+                  exit_on_error=false)
 
 Run the Julia unit tests listed in `tests`, which can be either a string or an array of
-strings, using `numcores` processors. (not exported)
+strings, using `numcores` processors. If `exit_on_error` is `false`, when one test
+fails, all remaining tests in other files will still be run; they are otherwise discarded,
+when `exit_on_error == true`.
 """
-function runtests(tests = ["all"], numcores = ceil(Int, Sys.CPU_CORES / 2))
+function runtests(tests = ["all"], numcores = ceil(Int, Sys.CPU_CORES / 2);
+                  exit_on_error=false)
     if isa(tests,AbstractString)
         tests = split(tests)
     end
+    exit_on_error && push!(tests, "--exit-on-error")
     ENV2 = copy(ENV)
     ENV2["JULIA_CPU_CORES"] = "$numcores"
     try
diff --git a/test/choosetests.jl b/test/choosetests.jl
index 56e42ab8a072c..b527dc8d5b216 100644
--- a/test/choosetests.jl
+++ b/test/choosetests.jl
@@ -2,7 +2,7 @@
 
 @doc """
 
-`tests, net_on = choosetests(choices)` selects a set of tests to be
+`tests, net_on, exit_on_error = choosetests(choices)` selects a set of tests to be
 run. `choices` should be a vector of test names; if empty or set to
 `["all"]`, all tests are selected.
 
@@ -10,8 +10,14 @@ This function also supports "test collections": specifically, "linalg"
  refers to collections of tests in the correspondingly-named
 directories.
 
-Upon return, `tests` is a vector of fully-expanded test names, and
-`net_on` is true if networking is available (required for some tests).
+Upon return, `tests` is a vector of fully-expanded test names,
+`net_on` is true if networking is available (required for some tests),
+and `exit_on_error` is true if an error in one test should cancel
+remaining tests to be run (otherwise, all tests are run unconditionally).
+
+Two options can be passed to `choosetests` by including a special token
+in the `choices` argument: "--skip", which makes all tests coming after
+be skipped, and "--exit-on-error" which sets the value of `exit_on_error`.
 """ ->
 function choosetests(choices = [])
     testnames = [
@@ -51,11 +57,14 @@ function choosetests(choices = [])
 
     tests = []
     skip_tests = []
+    exit_on_error = false
 
     for (i, t) in enumerate(choices)
         if t == "--skip"
             skip_tests = choices[i + 1:end]
             break
+        elseif t == "--exit-on-error"
+            exit_on_error = true
         else
             push!(tests, t)
         end
@@ -168,5 +177,5 @@ function choosetests(choices = [])
 
     filter!(x -> !(x in skip_tests), tests)
 
-    tests, net_on
+    tests, net_on, exit_on_error
 end
diff --git a/test/runtests.jl b/test/runtests.jl
index c2b8b4bfdc39c..927165bf4048b 100644
--- a/test/runtests.jl
+++ b/test/runtests.jl
@@ -4,7 +4,7 @@ using Base.Test
 include("choosetests.jl")
 include("testenv.jl")
 
-tests, net_on = choosetests(ARGS)
+tests, net_on, exit_on_error = choosetests(ARGS)
 tests = unique(tests)
 
 const max_worker_rss = if haskey(ENV, "JULIA_TEST_MAXRSS_MB")
@@ -33,6 +33,7 @@ cd(dirname(@__FILE__)) do
         n > 1 && addprocs_with_testenv(n)
         BLAS.set_num_threads(1)
     end
+    skipped = 0
 
     @everywhere include("testdefs.jl")
 
@@ -59,14 +60,18 @@ cd(dirname(@__FILE__)) do
                         resp = [e]
                     end
                     push!(results, (test, resp))
-                    if resp[1] isa Exception || resp[end] > max_worker_rss
+                    if resp[1] isa Exception
+                        if exit_on_error
+                            skipped = length(tests)
+                            empty!(tests)
+                        end
+                    elseif resp[end] > max_worker_rss
                         if n > 1
                             rmprocs(wrkr, waitfor=30)
                             p = addprocs_with_testenv(1)[1]
                             remotecall_fetch(include, p, "testdefs.jl")
-                        else
-                            # single process testing, bail if mem limit reached
-                            resp[1] isa Exception || error("Halting tests. Memory limit reached : $resp > $max_worker_rss")
+                        else # single process testing
+                            error("Halting tests. Memory limit reached : $resp > $max_worker_rss")
                         end
                     end
                     if !isa(resp[1], Exception)
@@ -181,7 +186,9 @@ cd(dirname(@__FILE__)) do
     if !o_ts.anynonpass
         println("    \033[32;1mSUCCESS\033[0m")
     else
-        println("    \033[31;1mFAILURE\033[0m")
+        println("    \033[31;1mFAILURE\033[0m\n")
+        skipped > 0 &&
+            println("$skipped test", skipped > 1 ? "s were" : " was", " skipped due to failure.\n")
         Base.Test.print_test_errors(o_ts)
         throw(Test.FallbackTestSetException("Test run finished with errors"))
     end