Through analysis of the pumping test data, which were affected by errors, an aquifer was investigated in order to present a new method for estimating the aquifer parameters. Based on the self-paced learning method in the machine learning field, a self-paced learning method based on the differential evolution algorithm was established and applied to the determination of aquifer parameters. This method was compared with other methods under errors of different levels. The numerical experimental results show that under errors of different levels, the differences between the values estimated with this method and traditional methods are small, and there are minor differences between the simulation data and original data. The results of the self-paced learning method, which was used for estimation of aquifer parameters based on pumping test data, are effective and reliable. The established algorithm is highly stable and it is seldom affected by data errors.