It is essential to thoroughly analyze the probability of a system failure of safety-critical systems before release. Since this is typically unfeasible through testing alone, probabilistic safety analysis of models is commonly used. However, many modeling languages are lacking in either expressiveness or the ability to reflect the system architecture. The results of these weaknesses are models that are inaccurate in terms of either their behavior or architecture with respect to the system being analyzed. The few languages that overcome both these hurdles are instead overly complicated, resulting in models that are difficult to understand and prone to unintentional errors. To overcome this issue, the modeling language PAFML (Pattern Assisted Failure Modeling Language) is here presented which, while being expressive and able to reflect the system architecture, still satisfies a high level of simplicity. PAFML models are analyzed through transformation to SSF (Stochastic StateFlow), a preexisting language for analysis of stochastic models. PAFML is evaluated both through an interview study with industry professionals in systems safety and through modeling several industrial systems and modeling patterns common in failure behavior models.