Using feedforwardnet with a linear transfer function

Is there a reason that the feedforwardnet architecture when used with the 'trainr' training function and a linear transfer function (ie purelin,poslin,satlin,etc.) produces 'NaN' in my results file? I have used 'trainr' with other transfer functions with success, and have paired trainr and a linear transfer function with the cascadeforwardnet architecture with success as well.