This paper presents an algorithm for the detection of single and multiple misfires in internal combustion engines, using only the signal of the knock sensor. Several experiments were conducted on a four-cylinder engine to obtain the necessary signals in the healthy and faulty conditions. The faulty conditions were created by disconnecting the wires of the spark-plugs. The signal of the knock sensor is analyzed and the time-domain signals of different orders of engine vibrations are extracted using a Vold-Kalman filter (VKF). A set of statistical features were then extracted from the time-domain signals of each order. These features are representative of the engine condition. In the next step, the computed features were plotted for different states and are analyzed. Consequently, a criterion for identifying the engine misfire is obtained based on the extracted features. The results show promising misfire detection capability of the proposed system, which can be used for engine misfire detection without mounting additional sensors.