diff --git a/AbstractRecurrent.lua b/AbstractRecurrent.lua
index 276ee73..7607808 100644
--- a/AbstractRecurrent.lua
+++ b/AbstractRecurrent.lua
@@ -55,6 +55,8 @@ end
 function AbstractRecurrent:updateGradInput(input, gradOutput)
    -- updateGradInput should be called in reverse order of time
    self.updateGradInputStep = self.updateGradInputStep or self.step
+   assert(self.updateGradInputStep >= self.step - self.rho,
+          string.format('Called backward more than rho+1=%d times', self.rho+1))
 
    -- BPTT for one time-step
    self.gradInput = self:_updateGradInput(input, gradOutput)
@@ -68,6 +70,8 @@ function AbstractRecurrent:accGradParameters(input, gradOutput, scale)
    -- accGradParameters should be called in reverse order of time
    assert(self.updateGradInputStep < self.step, "Missing updateGradInput")
    self.accGradParametersStep = self.accGradParametersStep or self.step
+   assert(self.accGradParametersStep >= self.step - self.rho,
+          string.format('Called backward more than rho+1=%d times', self.rho+1))
 
    -- BPTT for one time-step
    self:_accGradParameters(input, gradOutput, scale)
diff --git a/test/test.lua b/test/test.lua
index 62d33b0..374fc0a 100644
--- a/test/test.lua
+++ b/test/test.lua
@@ -640,6 +640,17 @@ function rnntest.Recurrent_old()
    mytester:assert(err < 0.0001, "Recurrent optim.checkgrad error")
 end
 
+function rnntest.RecurrentErrorOnExtraBackward()
+    local model = nn.Recurrent(
+        nn.Identity(), nn.Identity(), nil, nil, 1 --[[rho]])
+    local input = torch.rand(1)
+    model:training()
+    for i = 1, 3 do model:forward(input) end
+    for j = 1, 2 do model:backward(input, input) end
+    mytester:assertErrorPattern(function() model:backward(input, input) end,
+                                'Called backward more than rho%+1=2 times')
+end
+
 function rnntest.Recurrent()
    local batchSize = 4
    local dictSize = 100