The spark plug condition is an effective parameter on the combustion quality of a spark ignition (SI) engine. If the condition of the spark plug becomes abnormal, pollutions and the efficiency of the engine will be affected. In the present paper, a procedure is proposed based on the vibration analysis for the spark plug fault detection. Vibration signals of the SI engine were collected by an accelerometer under two spark plug conditions, namely, normal and abnormal conditions. In order to remove noises from signals, the wavelet denoising technique was used. Then, the feature extraction method by statistical parameters was applied to obtain fault-indicating information. In this work, seven feature parameters were employed in the feature extraction stage, namely, maximum, mean, standard deviations, the variance, the skewness, Kurtosis and impulse factors. The neural network (NN) was trained with seven neurons in the input layer. After constructing the optimum structure, the performance of the network was tested. Results showed that a high level of the efficiency was gained in the spark plug fault detection. Therefore, it can be mentioned that the proposed approach could reliably be used for the fault identification in the engine spark plug.