Accelerated life testing (ALT) is used to obtain failure data of products in a short time period and extrapolate the reliability of the products under normal operating conditions. Estimation of the reliability is based on models, which can be classified to parametric and non-parametric models. Non-parametric models are commonly used because of the distribution-free property. Proportional hazard model (PHM) and proportional odds model (POM) are two widely used non-parametric models for reliability prediction based on ALT data. These two models perform well if the underlying distributions are Weibull and log-logistic respectively. However, in some situations the test specimens may be obtained from a mixed population and using PHM or POM will result in inaccurate estimates of the reliability at normal operating conditions. In this paper, a proportional hazard-proportional odds (PH-PO) model is developed in order to obtain more accurate estimation for the failure time distributions of Weibull, log-logistic and a mixture of the two. This PH-PO model is based on a parameter family which makes PHM and POM special cases of the model. The performance of the PH-PO model is verified numerically using simulated data. The results show that in general the PH-PO model gives more accurate estimations of reliability.