diff --git a/CHANGELOG.rst b/CHANGELOG.rst
index 7eabf62..f8a9b7f 100644
--- a/CHANGELOG.rst
+++ b/CHANGELOG.rst
@@ -18,6 +18,7 @@ Changes:
 - Fix a bunch of warnings with new ``cffi`` versions and Python 3.6.
   `#14 <https://github.com/hynek/argon2_cffi/pull/14>`_
   `#16 <https://github.com/hynek/argon2_cffi/pull/16>`_
+- Add low-level bindings for Argon2id functions.
 
 
 ----
diff --git a/docs/api.rst b/docs/api.rst
index 4997f87..8686193 100644
--- a/docs/api.rst
+++ b/docs/api.rst
@@ -53,7 +53,7 @@ Low Level
 .. automodule:: argon2.low_level
 
 .. autoclass:: Type
-  :members: D, I
+  :members: D, I, ID
 
 .. autodata:: ARGON2_VERSION
 
diff --git a/docs/argon2.rst b/docs/argon2.rst
index b39cf01..33eed0e 100644
--- a/docs/argon2.rst
+++ b/docs/argon2.rst
@@ -12,7 +12,7 @@ It is designed to have both a configurable runtime as well as memory consumption
 
 This means that you can decide how long it takes to hash a password and how much memory is required.
 
-Argon2 comes in two variants:
+Argon2 comes in three variants:
 
 Argon2d
   is faster and uses data-depending memory access, which makes it less suitable for hashing secrets and more suitable for cryptocurrencies and applications with no threats from side-channel timing attacks.
@@ -21,6 +21,9 @@ Argon2i
   uses data-independent memory access, which is preferred for password hashing and password-based key derivation.
   Argon2i is slower as it makes more passes over the memory to protect from tradeoff attacks.
 
+Argon2id
+  is a hybrid of Argon2i and Argon2d, using a combination of data-depending and data-independent memory accesses, which gives some of Argon2i's resistance to side-channel cache timing attacks and much of Argon2d's resistance to GPU cracking attacks.
+
 
 Why “just use bcrypt” Is Not the Best Answer (Anymore)
 ------------------------------------------------------
diff --git a/src/argon2/_ffi_build.py b/src/argon2/_ffi_build.py
index 4a13cb9..cec4469 100644
--- a/src/argon2/_ffi_build.py
+++ b/src/argon2/_ffi_build.py
@@ -30,7 +30,11 @@
 )
 
 ffi.cdef("""\
-typedef enum Argon2_type { Argon2_d = ..., Argon2_i = ... } argon2_type;
+typedef enum Argon2_type {
+    Argon2_d = ...,
+    Argon2_i = ...,
+    Argon2_id = ...,
+} argon2_type;
 typedef enum Argon2_version {
     ARGON2_VERSION_10 = ...,
     ARGON2_VERSION_13 = ...,
diff --git a/src/argon2/low_level.py b/src/argon2/low_level.py
index 1272b20..53804a3 100644
--- a/src/argon2/low_level.py
+++ b/src/argon2/low_level.py
@@ -52,6 +52,15 @@ class Type(Enum):
     password hashing and password-based key derivation.  Argon2i is slower as
     it makes more passes over the memory to protect from tradeoff attacks.
     """
+    ID = lib.Argon2_id
+    r"""
+    Argon2\ **id** is a hybrid of Argon2i and Argon2d, using a combination of
+    data-depending and data-independent memory accesses, which gives some of
+    Argon2i's resistance to side-channel cache timing attacks and much of
+    Argon2d's resistance to GPU cracking attacks.
+
+    .. versionadded:: 16.3.0
+    """
 
 
 def hash_secret(secret, salt, time_cost, memory_cost, parallelism, hash_len,
diff --git a/tests/test_low_level.py b/tests/test_low_level.py
index 511f22e..7d8976b 100644
--- a/tests/test_low_level.py
+++ b/tests/test_low_level.py
@@ -39,6 +39,16 @@
 #           c29tZXNhbHQ$cZn5d+rFh+ZfuRhm2iGUGgcrW5YLeM6q7L3vBsdmFA0
 # 0.119 seconds
 # Verification ok
+#
+# Type:		Argon2id
+# Iterations:	2
+# Memory:		65536 KiB
+# Parallelism:	4
+# Hash:		1a9677b0afe81fda7b548895e7a1bfeb8668ffc19a530e37e088a668fab1c02a
+# Encoded:	$argon2id$v=19$m=65536,t=2,p=4$
+#           c29tZXNhbHQ$GpZ3sK/oH9p7VIiV56G/64Zo/8GaUw434IimaPqxwCo
+# 0.154 seconds
+# Verification ok
 
 TEST_HASH_I_OLD = (
     b"$argon2i$m=65536,t=2,p=4"
@@ -53,12 +63,19 @@
     b"$argon2d$v=19$m=65536,t=2,p=4$"
     b"c29tZXNhbHQ$cZn5d+rFh+ZfuRhm2iGUGgcrW5YLeM6q7L3vBsdmFA0"
 )
+TEST_HASH_ID = (
+    b"$argon2id$v=19$m=65536,t=2,p=4$"
+    b"c29tZXNhbHQ$GpZ3sK/oH9p7VIiV56G/64Zo/8GaUw434IimaPqxwCo"
+)
 TEST_RAW_I = binascii.unhexlify(
     b"20c8adf6a90550b08c03f5628b32f9edc9d32ce6b90e254cf5e330a40bcfc2be"
 )
 TEST_RAW_D = binascii.unhexlify(
     b"7199f977eac587e65fb91866da21941a072b5b960b78ceaaecbdef06c766140d"
 )
+TEST_RAW_ID = binascii.unhexlify(
+    b"1a9677b0afe81fda7b548895e7a1bfeb8668ffc19a530e37e088a668fab1c02a"
+)
 
 TEST_PASSWORD = b"password"
 TEST_SALT_LEN = 16
@@ -71,10 +88,12 @@
 i_and_d_encoded = pytest.mark.parametrize("type,hash", [
     (Type.I, TEST_HASH_I,),
     (Type.D, TEST_HASH_D,),
+    (Type.ID, TEST_HASH_ID,),
 ])
 i_and_d_raw = pytest.mark.parametrize("type,hash", [
     (Type.I, TEST_RAW_I,),
     (Type.D, TEST_RAW_D,),
+    (Type.ID, TEST_RAW_ID,),
 ])
 
 both_hash_funcs = pytest.mark.parametrize("func", [