The invention discloses a copy number variation detection method based on next generation sequencing. The method comprises the following steps: pre-processing copy number variation data, constructing a sliding window, calculating of statistics, implementing a replacement policy, constructing zero distribution, and carrying out performance evaluation of an algorithm. The performance evaluation of the algorithm comprises the steps of judging whether a relatively high correct positive rate can be acquired by the algorithm under the condition that a false positive rate is controllable, evaluating whether the algorithm can relatively accurately estimate a p value or not, detecting a boundary detection capability of copy number variation, and analyzing the calculation complexity of the algorithm. With the adoption of the copy number variation detection method, the problem of copy number variation detection errors, caused by the fact that sequencing platforms and sequencing levels are different, is solved, and a result is relatively accurate; data is normalized by utilizing characteristics of a multi-peak frequency histogram, so that a normal region and a copy number variation region are accurately divided; and a new model is established by a comprehensive effect of relevance between a variation reads number and a variation site, so that the inconsistency problem is solved, and the remarkable level of copy number variation is objectively estimated.