diff --git a/deps/gen_consts.jl b/deps/gen_consts.jl
index 0131d4e41..dd9cf5fe2 100644
--- a/deps/gen_consts.jl
+++ b/deps/gen_consts.jl
@@ -86,6 +86,10 @@ MPI_handle = [
 ]
 
 MPI_Cints = [
+    :MPI_THREAD_SINGLE,
+    :MPI_THREAD_FUNNELED,
+    :MPI_THREAD_SERIALIZED,
+    :MPI_THREAD_MULTIPLE,
     :MPI_PROC_NULL,
     :MPI_ANY_SOURCE,
     :MPI_ANY_TAG,
diff --git a/docs/src/environment.md b/docs/src/environment.md
index 1d9eeec8c..9f5103959 100644
--- a/docs/src/environment.md
+++ b/docs/src/environment.md
@@ -11,6 +11,7 @@ mpiexec
 ```@docs
 MPI.Abort
 MPI.Init
+MPI.Init_thread
 MPI.Initialized
 MPI.Finalize
 MPI.Finalized
diff --git a/src/MPI.jl b/src/MPI.jl
index 6f54a97aa..6b0b5cbe6 100644
--- a/src/MPI.jl
+++ b/src/MPI.jl
@@ -86,6 +86,14 @@ function __init__()
     ENV["UCX_MEM_MALLOC_RELOC"] = "no"
     ENV["UCX_MEM_EVENTS"] = "no"
 
+    # Julia multithreading uses SIGSEGV to sync thread
+    # https://docs.julialang.org/en/v1/devdocs/debuggingtips/#Dealing-with-signals-1
+    # By default, UCX will error if this occurs (issue #337)
+    if !haskey(ENV, "UCX_ERROR_SIGNALS")
+        # default is "SIGILL,SIGSEGV,SIGBUS,SIGFPE"
+        ENV["UCX_ERROR_SIGNALS"] = "SIGILL,SIGBUS,SIGFPE"
+    end
+    
     @require CuArrays="3a865a2d-5b23-5a0f-bc46-62713ec82fae" include("cuda.jl")
 end
 
diff --git a/src/consts/microsoftmpi.jl b/src/consts/microsoftmpi.jl
index 7c88ca52b..e4c76f71c 100644
--- a/src/consts/microsoftmpi.jl
+++ b/src/consts/microsoftmpi.jl
@@ -1,3 +1,5 @@
+# https://github.com/microsoft/Microsoft-MPI/blob/master/src/include/mpi.h
+
 const MPI_Aint   = Int
 const MPI_Offset = Int64
 const MPI_Count  = Int64
@@ -64,6 +66,10 @@ const MPI_C_DOUBLE_COMPLEX      = reinterpret(Cint, 0x4c001014)
 const MPI_File = Cint
 const MPI_FILE_NULL = Cint(0)
 
+const MPI_THREAD_SINGLE = Cint(0)
+const MPI_THREAD_FUNNELED = Cint(1)
+const MPI_THREAD_SERIALIZED = Cint(2)
+const MPI_THREAD_MULTIPLE = Cint(3)
 const MPI_PROC_NULL = Cint(-1)
 const MPI_ANY_SOURCE = Cint(-2)
 const MPI_ANY_TAG = Cint(-1)
diff --git a/src/consts/mpich.jl b/src/consts/mpich.jl
index 64dab1204..4636370fd 100644
--- a/src/consts/mpich.jl
+++ b/src/consts/mpich.jl
@@ -1,3 +1,5 @@
+# https://github.com/pmodels/mpich/blob/master/src/include/mpi.h.in
+
 const MPI_Aint = Int
 const MPI_Count = Int64
 const MPI_Offset = Int64
@@ -69,6 +71,10 @@ const MPI_UINT64_T = Cint(1275070526)
 const MPI_C_FLOAT_COMPLEX = Cint(1275070528)
 const MPI_C_DOUBLE_COMPLEX = Cint(1275072577)
 
+const MPI_THREAD_SINGLE = Cint(0)
+const MPI_THREAD_FUNNELED = Cint(1)
+const MPI_THREAD_SERIALIZED = Cint(2)
+const MPI_THREAD_MULTIPLE = Cint(3)
 const MPI_PROC_NULL = Cint(-1)
 const MPI_ANY_SOURCE = Cint(-2)
 const MPI_ANY_TAG = Cint(-1)
diff --git a/src/consts/openmpi.jl b/src/consts/openmpi.jl
index 8419bc8cb..0dd871a33 100644
--- a/src/consts/openmpi.jl
+++ b/src/consts/openmpi.jl
@@ -1,3 +1,5 @@
+# https://github.com/open-mpi/ompi/blob/master/ompi/include/mpi.h.in
+
 const MPI_Aint = Int
 const MPI_Count = Int64
 const MPI_Offset = Int64
@@ -75,6 +77,10 @@ const MPI_UINT64_T = Cint(65)
 const MPI_C_FLOAT_COMPLEX = Cint(69)
 const MPI_C_DOUBLE_COMPLEX = Cint(70)
 
+const MPI_THREAD_SINGLE = Cint(0)
+const MPI_THREAD_FUNNELED = Cint(1)
+const MPI_THREAD_SERIALIZED = Cint(2)
+const MPI_THREAD_MULTIPLE = Cint(3)
 const MPI_PROC_NULL = Cint(-2)
 const MPI_ANY_SOURCE = Cint(-1)
 const MPI_ANY_TAG = Cint(-1)
diff --git a/src/environment.jl b/src/environment.jl
index 6e2d0f9d0..a7bb3f075 100644
--- a/src/environment.jl
+++ b/src/environment.jl
@@ -34,9 +34,11 @@ end
 
 Initialize MPI in the current process.
 
-All MPI programs must contain exactly one call to `MPI.Init()`. In particular, note that it is not valid to call `MPI.Init` again after calling [`MPI.Finalize`](@ref).
+All MPI programs must contain exactly one call to `MPI.Init` or
+[`MPI.Init_thread`](@ref). In particular, note that it is not valid to call `MPI.Init` or
+`MPI.Init_thread` again after calling [`MPI.Finalize`](@ref).
 
-The only MPI functions that may be called before `MPI.Init()` are
+The only MPI functions that may be called before `MPI.Init`/`MPI.Init_thread` are
 [`MPI.Initialized`](@ref) and [`MPI.Finalized`](@ref).
 
 # External links
@@ -53,6 +55,64 @@ function Init()
     end
 end
 
+@enum ThreadLevel begin
+    THREAD_SINGLE     = MPI_THREAD_SINGLE
+    THREAD_FUNNELED   = MPI_THREAD_FUNNELED
+    THREAD_SERIALIZED = MPI_THREAD_SERIALIZED
+    THREAD_MULTIPLE   = MPI_THREAD_MULTIPLE
+end
+    
+
+"""
+    Init_thread(required::ThreadLevel)
+
+Initialize MPI and the MPI thread environment in the current process. The argument
+specifies the required thread level, which is one of the following:
+
+ - `MPI.THREAD_SINGLE`: Only one thread will execute.
+ - `MPI.THREAD_FUNNELED`: The process may be multi-threaded, but the application must ensure that only the main thread makes MPI calls.
+ - `MPI.THREAD_SERIALIZED`: The process may be multi-threaded, and multiple threads may make MPI calls, but only one at a time (i.e. all MPI calls are serialized).
+ - `MPI.THREAD_MULTIPLE`: Multiple threads may call MPI, with no restrictions.
+
+Tne function will return the provided `ThreadLevel`, and values may be compared via inequalities, i.e.
+```julia
+if Init_thread(required) < required
+    error("Insufficient threading")
+end
+```
+
+All MPI programs must contain exactly one call to [`MPI.Init`](@ref) or
+`MPI.Init_thread`. In particular, note that it is not valid to call `MPI.Init` or
+`MPI.Init_thread` again after calling [`MPI.Finalize`](@ref).
+
+The only MPI functions that may be called before `MPI.Init`/`MPI.Init_thread` are
+[`MPI.Initialized`](@ref) and [`MPI.Finalized`](@ref).
+
+# External links
+$(_doc_external("MPI_Init_thread"))
+"""
+function Init_thread(required::ThreadLevel)
+    REFCOUNT[] == -1 || error("MPI.REFCOUNT in incorrect state: MPI may only be initialized once per session.")
+    r_provided = Ref{ThreadLevel}()
+    # int MPI_Init_thread(int *argc, char ***argv, int required, int *provided)
+    @mpichk ccall((:MPI_Init_thread, libmpi), Cint,
+                  (Ptr{Cint},Ptr{Cvoid}, ThreadLevel, Ref{ThreadLevel}),
+                  C_NULL, C_NULL, required, r_provided)
+    provided = r_provided[]
+    if provided < required
+        @warn "Thread level requested = $required, provided = $provided"
+    end
+    
+    REFCOUNT[] = 1
+    atexit(refcount_dec)
+
+    for f in mpi_init_hooks
+        f()
+    end
+    return provided
+end
+
+
 """
     Finalize()
 
diff --git a/test/runtests.jl b/test/runtests.jl
index e0e6a4128..ac2c08224 100644
--- a/test/runtests.jl
+++ b/test/runtests.jl
@@ -17,9 +17,6 @@ const coverage_opts =
                       JL_LOG_USER => "user",
                       JL_LOG_ALL => "all")
 
-# Files to run with mpiexec -n 1
-singlefiles = ["test_spawn.jl"]
-
 function runtests()
     nprocs = clamp(Sys.CPU_THREADS, 2, 4)
     exename = joinpath(Sys.BINDIR, Base.julia_exename())
@@ -33,8 +30,12 @@ function runtests()
     for f in testfiles
         coverage_opt = coverage_opts[Base.JLOptions().code_coverage]
         mpiexec() do cmd
-            if f ∈ singlefiles
+            if f == "test_spawn.jl"
                 run(`$cmd -n 1 $exename --code-coverage=$coverage_opt $(joinpath(testdir, f))`)
+            elseif f == "test_threads.jl"
+                withenv("JULIA_NUM_THREAD" => "4") do
+                    run(`$cmd -n $nprocs $exename --code-coverage=$coverage_opt $(joinpath(testdir, f))`)
+                end
             else
                 run(`$cmd -n $nprocs $exename --code-coverage=$coverage_opt $(joinpath(testdir, f))`)
             end
diff --git a/test/test_threads.jl b/test/test_threads.jl
new file mode 100644
index 000000000..78e0086d7
--- /dev/null
+++ b/test/test_threads.jl
@@ -0,0 +1,38 @@
+using Test, Pkg
+using MPI
+
+if get(ENV,"JULIA_MPI_TEST_ARRAYTYPE","") == "CuArray"
+    using CuArrays
+    ArrayType = CuArray
+else
+    ArrayType = Array
+end
+
+provided = MPI.Init_thread(MPI.THREAD_MULTIPLE)
+
+comm = MPI.COMM_WORLD
+size = MPI.Comm_size(comm)
+rank = MPI.Comm_rank(comm)
+
+const N = 10
+
+dst = mod(rank+1, size)
+src = mod(rank-1, size)
+
+if provided == MPI.THREAD_MULTIPLE
+    send_arr = collect(1.0:N)
+    recv_arr = zeros(N)
+
+    reqs = Array{MPI.Request}(undef, 2N)
+    
+    Threads.@threads for i = 1:N
+        reqs[N+i] = MPI.Irecv!(@view(recv_arr[i:i]), src, i, comm)
+        reqs[i] = MPI.Isend(@view(send_arr[i:i]), dst, i, comm)
+    end
+
+    MPI.Waitall!(reqs)
+
+    @test recv_arr == send_arr
+end        
+
+MPI.Finalize()