The quality evaluation of vehicle sounds and vibrations is a multidimensional phenomenon. An important quality dimension is sportiness. This study develops a model to predict the sportiness of vehicle engine start sounds based on signal properties. The driving condition 'engine start' exhibits a non-stationary complex character. In this driving condition, the relative contributions of various sound sources interact with each other in time and frequency domains. According to this interaction and the individual elements of non-stationary signals, vehicle sounds evoke high or low sportiness. The development of the model consists of the determination of the sportiness judgements and the investigation of the relationship between the sportiness judgements and the signal properties. First, the binaurally recorded sounds of 38 cars from different brands with different motorisation and their modified versions were presented to the subjects and they indicated the intensity of their sportiness association on a quasi-continuous scale. Second, the signal properties of the recorded and synthesised sounds were analysed. Third, a prediction model was developed based on the sportiness judgements and the signal properties. The knowledge, taste, interpretation and expectations of vehicle customers vary significantly. Therefore, the requirement of model adaptation for different customer groups was discussed.