This paper presents two error models to evaluate safety of a software error detection method. The proposed models analyze the impact on program overhead in terms of memory code area and increased execution time when the studied error detection technique is applied. For faults affecting the processor's registers, analytic formulas are derived to estimate the failure rate before program execution. These formulas are based on probabilistic methods and use statistics of the program, which are collected during compilation. The studied error detection technique was applied to several benchmark programs and then program overhead and failure rate was estimated. Experimental results validate the estimated performances and show the effectiveness of the proposed evaluation formulas.