diff --git a/NAMESPACE b/NAMESPACE
index 90923cbfe..fed462848 100644
--- a/NAMESPACE
+++ b/NAMESPACE
@@ -75,6 +75,7 @@ export(image_load)
 export(image_to_array)
 export(imagenet_decode_predictions)
 export(imagenet_preprocess_input)
+export(implementation)
 export(inception_v3_preprocess_input)
 export(initializer_constant)
 export(initializer_glorot_normal)
diff --git a/R/utils.R b/R/utils.R
index 0c51f1b74..64aa42d4e 100644
--- a/R/utils.R
+++ b/R/utils.R
@@ -109,6 +109,28 @@ normalize <- function(x, axis = -1, order = 2) {
 }
 
 
+#' Keras implementation
+#' 
+#' Obtain a reference to the Python module used for the implementation of Keras.
+#' 
+#' There are currently two Python modules which implement Keras:
+#' 
+#' - keras ("keras")
+#' - tensorflow.contrib.keras ("tensorflow")
+#' 
+#' This function returns a reference to the implementation being currently 
+#' used by the keras package. The default implementation is "tensorflow".
+#' You can override this by setting the `KERAS_IMPLEMENTATION` environment
+#' variable to "keras".
+#' 
+#' @return Reference to the Python module used for the implementation of Keras.
+#' 
+#' @export
+implementation <- function() {
+  keras
+}
+
+
 #' Keras backend tensor engine
 #' 
 #' Obtain a reference to the `keras.backend` Python module used to implement
diff --git a/man/implementation.Rd b/man/implementation.Rd
new file mode 100644
index 000000000..373551f52
--- /dev/null
+++ b/man/implementation.Rd
@@ -0,0 +1,26 @@
+% Generated by roxygen2: do not edit by hand
+% Please edit documentation in R/utils.R
+\name{implementation}
+\alias{implementation}
+\title{Keras implementation}
+\usage{
+implementation()
+}
+\value{
+Reference to the Python module used for the implementation of Keras.
+}
+\description{
+Obtain a reference to the Python module used for the implementation of Keras.
+}
+\details{
+There are currently two Python modules which implement Keras:
+\itemize{
+\item keras ("keras")
+\item tensorflow.contrib.keras ("tensorflow")
+}
+
+This function returns a reference to the implementation being currently
+used by the keras package. The default implementation is "tensorflow".
+You can override this by setting the \code{KERAS_IMPLEMENTATION} environment
+variable to "keras".
+}