In the tight reservoir like mahu with strong heterogeneity, large differences in the productivity of oil wells and obvious differences in the decline characteristics, the production decline analysis and prediction have important guiding significance to the management of oil well production system and the preparation and optimization of oil field development plan, and are also an important part of reservoir performance analysis. However, the traditional empirical decline models used today, such as the Arps production decline model, power law exponential decline model, Valkó extended exponential decline model, and Duong decline model, have different applicability. When using these methods to fit the production data of the fractured horizontal wells in the Mahu tight reservoir, the accuracy of the fitted results is low. Therefore, a new method of combination that uses different models for fitting in stages based on the model fitting results of Mahu oil wells in different stages is explored, and the examples are applied to verify it in this block. The research results show that this new method has lower error rate and higher degree of agreement than the traditional ones. In addition, by using this method, the prediction results of Estimated Ultimate Recovery（EUR） for the tight oil wells in Mahu reservoir are more accurate, and the method is highly feasible, which can be used as a reference for the research on the production decline law and the productivity prediction of other similar tight oil reservoirs.