Traditional reliability analysis and prediction are performed by utilizing observed failure or degradation data of test units or field observations. Reliability testing is usually performed to predict reliability or performed as acceptance testing or reliability demonstration test. Moreover, in many cases, reliability tests are performed repeatedly during the entire life of the system by testing different samples with different characteristics in the system. At the end of each test, available data only show the number and combination of failed units which are then used for reliability estimation and prediction. This paper investigates several effective approaches to obtain expressions for system reliability metrics under different scenarios, considering the mixture characteristics of the units. The proposed approaches apply to general cases when the population size, as well as the mixture of units, increase over time. A simulation model is utilized to validate the proposed models.