This paper studies the detection of the nonlinearity of time series sampled from continuous dynamics systems by using the surrogate data method. The results show that under the different sampling conditions, the detection finds different nonlinearity of chaotic time series. Especially for the oversampling time series, there can often be some illusive results. For this, we suggest that it is best to apply nonlinear values as testing statistics for detecting nonlinearity of oversampling time series.