Abstract: Vehicle driving states that vary with time were taken as a Markov process to reduce the influences of uncertainty and small variations in driving speed on the driving cycle model. Collected data were classified into model events of idling, acceleration, deceleration, and constant velocity using maximum likelihood estimation. The model events with similar average speeds were categorized into six states, and their transition probabilities were calculated. Pseudo-random numbers satisfying distribution of the state transition probabilities were generated to extend the length of driving cycle. The application of the driving cycle model to the roads in Hefei (a city in China) shows that the average error in obtained typical driving cycles was 7.81% compared with the experimental data in terms of main driving characteristic parameters of a typical driving cycle，decreasing by 14.72% compared with that by principal component analysis.