Mostly, defect correlation is caused by the defect detected capability masked by other defects. Defect correlation affects the result of the test and makes the results in accurate and distorts the estimated results of software reliability assessment models. From the standpoint of defects themselves, this paper makes a detailed analysis of defect correlation and gives the reasons of the software testing and reliability assessment out of action. By applying the generalized correlation to improve the existing reliability assessment models, this paper puts forward the P-NHPP (phase-nonhomogeneous poisson process) reliability model to make the reliability assessment parameters more coordinated with the actual defect number. Experimental results show that P-NHPP is better and has a fairly accurate prediction capability.