A complex fault-block reservoir usually has large layer span, many thin layers and oil-water systems. The available data are limited in the exploration stage. One of the biggest advantages is that each well is tested several times. Based on comprehensive geological study, this paper proposed a well testing analysis methodology to carry out reservoir evaluation, and the results are applied to the oilfield development plan. Well testing can reflect the static and dynamic characteristics of reservoir at the same time. It shows high application value in the exploration stage. The analysis methodology of well testing in complex fault-block oilfield mainly includes three aspects: (1) Establish a novel calculation method to obtain crude oil parameters of all tested intervals, which are necessary in well test interpretation. (2) Build reliable interpretation models to evaluate wellbore, formation property, productivity and boundary according to pressure derivatives curves. (3) Propose a concise evaluation process to show reservoir characteristics in both vertical and plane. This methodology is applied in Doseo Oilfield and proved to be a great methodology in complex fault-block reservoirs. The calculated crude oil parameters are good matched with limited PVT data. The errors are less than 10%. The formation characteristics obtained from the models including permeability, faults and aquifer energy are well verified by seismic and well-logging interpretation. The reservoirs in middle area on the plane and Zone K in the vertical show high productivity and permeability. They can be selected as the key oil regions for the first production. Similar formation characteristics are reflected by the analogous shape of pressure derivatives. They can be put into production with the same development methods to obtain higher benefits. The innovation is that this paper proposes an analysis methodology of well testing to calculate right crude oil parameters, identify formation characteristics and key oil regions, and find similar reservoirs. It can decrease the errors which are caused by parameter uncertainties and multi-solution of interpretation models.