By progress of electronic technology and its vast use in automotive industry, automotive suppliers and other related organizations have agreed on OBD standard to reach more coordination in diagnosis. Engine misfire monitoring is one of the most difficult rules in this standard. Since misfire causes severe pollution of engines and it causes mechanical damages to engine parts, so preventing it is very important. According to standards, attention to misfire detection mechanisms should be provided in motor controllers, to satisfy the OBD II rules. In this thesis, engine combustion process has been modeled as a nonlinear discrete dynamic system. In this model, the event of normal firing or misfire is considered as an input to it, and the crankshaft speed fluctuation as a system output response. The current crankshaft speed is related to the current and previous firing events and also the previous speed. Once an inverse model from the crankshaft speed fluctuation signal to the firing event signal to be identified, the engine misfire can be detected more accurately, because the output signal has a higher signal to noise ratio for the misfire signature than its input signal. For system identification, a two layer dynamic neural network is used and since the previous firing events have influence on current firing event signal, a feedback path is described from output to input. A backpropagation learning rule is used for the network’s training. Low sampling rate of data is the advantage of this method, that is, one data point per firing event. This means one data point for a 180 degree rotation of the crankshaft in a 4-cylinder engine that easily acquire from engine control unit. The data consist of the firing event signal, the crankshaft speed fluctuation, the average of crankshaft speed and the average of manifold absolute pressure are acquired in different engine running condition and according to this data, training of network is done. According to the results that have been obtained in simulations, the network could detect misfire well. In comparison with other strategies that need to high sampling rate, this is a suitable method for misfire detection.