BACKGROUND: Several statistical methods have been applied to detect signals in spontaneous reporting databases. The Thailand Food and Drug Administration (Thai FDA) uses the reporting odds ratio (ROR) method for signal detection because of its ease of implementation and interpretation. The performance of different methods needs to be explored to determine need for modifications to the Thai FDA signal detection system. OBJECTIVES: To examine the concordance between the ROR and the Bayesian Confidence Propagation Neural Network (BCPNN) methods in identifying adverse drug reaction (ADR) signals using the Thai FDA database. METHODS: The two methods were retrospectively applied to identify ADRs reported with antiretroviral (ARV) drugs using the dataset from 1990 to 2006. The criteria of lower limit of 95% confidence interval of ROR > 1, 3 or more cases; and information component (IC) 2 standard deviations > 0, were used to identify signals for ROR and BCPNN, respectively. The sensitivity, specificity and agreement of timing of signal detection were measured for the concordance of the ROR in respect to the BCPNN. RESULTS: Using the BCPNN as a reference method, the ROR has high sensitivity and specificity. For the agreement of timing of signal detection between the ROR and the BCPNN, signals detected by both methods were in agreement on the first time of signal detection for 76.92%. CONCLUSIONS: ROR and BCPNN are comparable in identify signals of potential ADRs. Comparisons using other drug classes will provide additional insight into the performance of these two methods.