Many non-homogeneous Poisson process software reliability growth models are characterized by their mean value functions. Mean value functions of coverage-based models are usually obtained as composite functions of the coverage growth function and the function relating the number of detected faults to the coverage. This paper performs empirical evaluation of the relationships between the number of detected faults and the coverage embedded in the coverage- based software reliability growth models. It is also illustrated that integration of well-performing coverage growth functions and relationships between the number of detected faults and the coverage produces well-performing mean value functions.