We present an assessment method of automobile exhaust using three-dimensional spectrogram features to solve the nonlinear and time-varying problems of automobile exhaust emissions. The method takes advantage of spectral analysis to obtain the three-dimensional spectrogram features of automobile exhaust. These three-dimensional spectrogram features, being considered as the input variables, are fed to radial basis function neural network (RBFNN) adapting the K-means algorithm. After completing the training of RBFNN, the three-dimensional spectrograms, being unseen by the network before, are fed to the well-trained network for testing, which achieves the assessment level of automobile exhaust via the classification of three-dimensional spectrogram features. The numerical experiments indicate that the proposed method has a high accuracy.