This study proposes a new method to evaluate a probability of detection (POD) as a function of more than one flaw parameter. The main idea of the method is to assume that signals due to flaws described with a given parameter x have a normal distribution of a mean of μ( x) and a standard deviation of σ ( x), and to use a combination of signals calculated by numerical simulations and experimental data to evaluate μ( x) and σ ( x), respectively. The method does not postulate a closed-form of μ( x) found in conventional approaches, and it evaluates a few parameters that characterize the distribution using maximum likelihood analysis to calculate POD. This allows POD evaluation for data that does not satisfy linearity or constant variance assumptions without transformation. The proposed method is demonstrated through analyzing simulated eddy current signals due to flaws appearing in type 316L stainless steel welds. The results of the demonstration confirm that the proposed method can provide the POD with its confidence bounds as a function of the depth and the length of a flaw. The results also showed that the proposed method does not require a large amount of experimental data compared to conventionalâ vs. a analysis.