Traffic flow time series at different time scales have different fluctuation characteristics. In order to analyze the discontinuity characteristics of traffic flow, a traffic volume time series is adopted as an empiric study objective. The multi-scale sharp variation points of the sequence are obtained by using wavelet transform. The scales are divided based on the corresponding number of the zero-crossing points of wavelet transform, and the hierarchy of the scales is analyzed. The results of calculation indicate that the number of sharp variation points of the traffic flow on a mutation scale is self-similar in a certain scale if the number of the sharp variation points is used as a measure. So, sharp variations of the traffic flow in different scales are scale-free.