The invention provides a discontinuous context modeling and maximum entropy principle based gene compression method. The method includes that in a first stage, two actual situations with reference sequences and without reference sequences are considered simultaneously, and repetitive sequences in or among gene sequences are represented with a dictionary method, so that compression efficiency is improved; in a second stage, a statistical encoder is composed of a predictor and an arithmetic encoder, the predictor introduces a modeling scheme in a discontinuous bit context combination type on the basis of traditional continuous context models for non-repetitive sequences, and prediction probabilities independently generated by all models are merged with a Logistic regression formula generated on the basis of the maximum entropy principle, so that a more accurate final prediction probability value is obtained and put into the arithmetic encoder. By the method, compression efficiency can be improved remarkably, and efficient storage is realized.