Model-based software development using MATLAB Simulink is widely used in safety-critical domains. The reliability properties of the developed software have to be numerically evaluated for the precise system-level dependability analysis. Data errors occurred in RAM or CPU registers can propagate to critical outputs and cause a failure. The reliability properties can be evaluated at the assembly level, i.e. on the compiled instructions, by performing a probabilistic modeling of data errors. It is more accurate to conduct reliability assessment at the low level, however, the method scalability is questionable due to the complicated procedure, complexity of the assembly code, and considerable computation effort. Thus assembly-level evaluation is unsuitable for huge and complex Simulink models. In addition, it is more convenient for design engineers to estimate dependability properties of Simulink models and even to design reliable control systems at the model level. In this paper, we propose a method for the reliability evaluation of Simulink models at the model level, extended with the assembly-level evaluation. More specifically, we transform the Simulink model into a stochastic dual-graph error propagation model and specify the reliability properties of individual Simulink blocks by loading the data from a database that have been obtained via the assembly-level evaluation. We verified the efficiency of the proposed method by the comparison of the reliability properties, evaluated at the assembly level and at the model level. The experimental results indicate that the reliability metrics, evaluated at the model level, are almost equivalent to the ones, evaluated at the assembly level. Most prominently, the application of the proposed model-level assessment can reduce the computation and engineering effort, and increase the method scalability.