The software reliability testing has many disadvantages in practice, such as high complexity of constructing operational profiles and poor fault detection efficiency. Oppositely, the directed testing with a high fault detection rate is incapable of estimating reliability quantificationally. To solve this problem, a hybrid testing combining reliability and directed testing as well as a reliability model based on the order statistic (OS) model were presented by Mitchell. An extended research on Mitchell’s work is proposed. Firstly, the most proper distribution of the fault’s failure rate which tends to be lognormal is suggested, and a detailed form of the OS model based on lognormal and the corresponding parameter estimation method are proposed, respectively. Secondly, an implementing framework for the hybrid testing is proposed. Finally, the hybrid testing and the OS model are applied on a real website system. The experimental results indicate: the hybrid testing has more efficient fault detection power and lower testing cost than the reliability testing; compared with three traditional software reliability growth models, the OS model has a best or pretty estimation or prediction power for each data set; and for the failure data set collected from hybrid testing, the OS model also achieves an acceptable estimation result.