Compositionality is a key property in the development and analy-
sis of component-based systems. In non-probabilistic formalisms such
as Labelled Transition Systems (LTS) the functional behaviour of a
system can be readily constructed from the behaviours of its parts.
However, this is not true for probabilistic extensions of LTS, which
are necessary to analyse non-functional properties such as reliability.
We propose Probabilistic Component Automata (PCA) as a proba-
bilistic extension to Interface Automata to automatically construct a
system model by composing models of its sub-components. In par-
ticular, we focus on modelling failure scenarios, failure handling and
failure propagation. Additionally, we propose a novel algorithm based
on Compositional Reachability Analysis to mitigate the well-known
state-explosion problem associated with composable models. Both
Probabilistic Component Automata and the reduction algorithm have
been implemented in the LTSA tool.