diff --git a/src/GraphNeuralNetworks.jl b/src/GraphNeuralNetworks.jl
index 86bab1896..011763160 100644
--- a/src/GraphNeuralNetworks.jl
+++ b/src/GraphNeuralNetworks.jl
@@ -64,6 +64,7 @@ export
     SGConv,
     
     # layers/pool
+    ConcatPool,
     GlobalPool,
     GlobalAttentionPool,
     TopKPool,
diff --git a/src/layers/pool.jl b/src/layers/pool.jl
index 88ed3ddd0..c9c6e36c8 100644
--- a/src/layers/pool.jl
+++ b/src/layers/pool.jl
@@ -1,5 +1,40 @@
 using DataStructures: nlargest
 
+@doc raw"""
+    ConcatPool(pooling_layer)
+
+```math
+\mathbf{x}_i' = [\mathbf{x}_i; \mathbf{u}_V]
+```
+
+# Arguments
+
+- `pooling_layer`: 
+
+# Examples
+
+```julia
+using Flux, GraphNeuralNetworks, Graphs
+
+add_pool = ConcatPool(GlobalPool(mean))
+
+g = GNNGraph(rand_graph(10, 4))
+X = rand(32, 10)
+pool(g, X) # => 64x10 matrix
+```
+"""
+struct ConcatPool <: GNNLayer
+    pool::GNNLayer
+end
+  
+function (l::ConcatPool)(g::GNNGraph, x::AbstractArray)
+    g_feat = applylayer(l.pool, g, x)
+    feat_arr = broadcast_nodes(g, g_feat)
+    return vcat(x, feat_arr)
+end
+
+(l::ConcatPool)(g::GNNGraph) = GNNGraph(g, gdata=l(g, node_features(g)))
+
 @doc raw"""
     GlobalPool(aggr)
 
@@ -44,7 +79,6 @@ end
 
 (l::GlobalPool)(g::GNNGraph) = GNNGraph(g, gdata=l(g, node_features(g)))
 
-
 @doc raw"""
     GlobalAttentionPool(fgate, ffeat=identity)
 
diff --git a/test/layers/pool.jl b/test/layers/pool.jl
index f7bb74a83..6d84b01d4 100644
--- a/test/layers/pool.jl
+++ b/test/layers/pool.jl
@@ -21,6 +21,35 @@
         test_layer(p, g, rtol=1e-5, exclude_grad_fields = [:aggr], outtype=:graph)
     end
 
+    @testset "ConcatPool" begin
+        p = GlobalPool(+)
+        l = ConcatPool(p)
+        n = 10
+        chin = 6
+        X = rand(Float32, chin, n)
+        g = GNNGraph(random_regular_graph(n, 4), ndata=X, graph_type=GRAPH_T)
+        y = p(g, X)
+        u = l(g, X)
+
+        @test size(u) == (chin*2, n)
+        @test u[1:chin,:] ≈ X
+        @test u[chin+1:end,:] ≈ repeat(y, 1, n)
+
+        n = [5, 6, 7]
+        ng = length(n)
+        g = Flux.batch([GNNGraph(random_regular_graph(n[i], 4), 
+                                 ndata=rand(Float32, chin, n[i]),
+                                 graph_type=GRAPH_T) 
+                        for i=1:ng])
+        y = p(g, g.ndata.x)
+        u = l(g, g.ndata.x)
+        @test size(u) == (chin*2, sum(n))
+        @test u[1:chin,:] ≈ g.ndata.x
+        @test u[chin+1:end,:] ≈ hcat([repeat(y[:,i], 1, n[i]) for i=1:ng]...)
+      
+        test_layer(p, g, rtol=1e-5, exclude_grad_fields = [:aggr], outtype=:graph)
+    end
+
     @testset "GlobalAttentionPool" begin
         n = 10
         chin = 6