Reliability testing is an indispensable tool for evaluating the lifetime of a product. To satisfy the needs for the use of advanced technologies, the requirements of design, manufacturing and reliability for some key components are becoming higher and higher. Because of the high reliability of such components, it is difficult to obtain failure time data in a limited time period via regular life tests or even accelerated life test (ALT). As an alternative, accelerated degradation test (ADT) can also be conducted to collect performance degradation data under accelerated operations. The issue is how to evaluate the reliability of the product using both ALT and ADT data. To solve the problem, a multi data fusion model needs to be developed to integrate these two types of data. In this paper, an Expectation-maximization (EM) algorithm is developed for the estimation of model parameters in data fusion. The method evaluates the product's reliability more accurately. A numerical example is presented to illustrate the use of the proposed method in practice, and show the value of the method in improving the estimation accuracy.