In this thesis, we investigate reliable modelling within a stochastic process algebra framework. Primarily, we consider issues of variance in stochastic process algebras as a measure of model reliability. This is in contrast to previous research in the field which has tended to centre around mean behaviour and steady-state solutions. We present a method of stochastic aggregation for analysing generally-distributed processes. This allows us more descriptive power in representing stochastic systems and thus gives us the ability to create more accurate models. We improve upon two well-developed Markovian process algebras and show how their simpler paradigm can be brought to bear on more realistic synchronisation models. Now, reliable performance figures can be obtained for systems, where previously only approximations of unknown accuracy were possible. Finally, we describe reliability definitions and variance metrics in stochastic models and demonstrate how systems can be made more reliable through careful combination under stochastic process algebra operators.