We have developed a method for estimating software reliability of
a program using a Markov reward model. The method uses an op-erational profile of a program, and the estimated reliability of each
module, to estimate the reliability of a program. UsingiLTL,we
show how a variety of reliability properties may be specified and
we provide an algorithm for checking these properties. While our
technique provides a promising method for rigorous compositional
software reliability estimation, empirical studies with real software
systems remain to be carried out. Moreover, further research is
needed to quantify the effect of deviations from the assumptions
used in our model.