A Bayesian neural network (BCPNN) method is introduced for detecting adverse drug reaction (ADR) signals from a large database of ADR reports. The database, maintained by the Uppsala Monitoring Centre on behalf of the World Health Organization (WHO), contains nearly two million reports, with about 35,000 new reports added quarterly. The task of identifying new ADR signals is challenging due to the large volume of data. The BCPNN is a flexible, automated method that can handle large datasets, incomplete data, and complex variables. It uses information theory to identify drug-ADR combinations that are highly associated compared to the general data. The method is transparent and flexible, allowing for different types of searches. The BCPNN was tested on quarterly updates and found to detect 1004 suspected drug-ADR combinations at the 97.5% confidence level. Of these, 307 were potentially serious ADRs, 53 related to new drugs, and 12 were not recorded in major pharmacological references. The results indicate that the BCPNN can effectively detect significant signals from the WHO database. It is an important tool for improving the detection of ADR signals from large numbers of spontaneously reported cases. The WHO database contains nearly two million case reports of suspected ADRs for anonymous patients. Each report includes administrative data, patient data, ADR data, medication data, and other information. The database has 49 fields, though not all are filled in each report. The BCPNN is a feed-forward neural network that uses Bayes' law for learning and inference. It is self-organizing, suitable for parallel computers, and provides an efficient computational model. The BCPNN can handle missing data and is transparent, allowing for easy validation and checking. It is also efficient in training and searching through the database using a sparse matrix method. The Bayesian approach to signal generation involves calculating the probability of an ADR occurring with a specific drug. The BCPNN uses Bayes' law to calculate the probability of an ADR given a drug, and the probability of a drug given an ADR. This allows for the identification of drug-ADR combinations that are more likely to be associated than expected. The BCPNN is a powerful tool for detecting ADR signals from large datasets.A Bayesian neural network (BCPNN) method is introduced for detecting adverse drug reaction (ADR) signals from a large database of ADR reports. The database, maintained by the Uppsala Monitoring Centre on behalf of the World Health Organization (WHO), contains nearly two million reports, with about 35,000 new reports added quarterly. The task of identifying new ADR signals is challenging due to the large volume of data. The BCPNN is a flexible, automated method that can handle large datasets, incomplete data, and complex variables. It uses information theory to identify drug-ADR combinations that are highly associated compared to the general data. The method is transparent and flexible, allowing for different types of searches. The BCPNN was tested on quarterly updates and found to detect 1004 suspected drug-ADR combinations at the 97.5% confidence level. Of these, 307 were potentially serious ADRs, 53 related to new drugs, and 12 were not recorded in major pharmacological references. The results indicate that the BCPNN can effectively detect significant signals from the WHO database. It is an important tool for improving the detection of ADR signals from large numbers of spontaneously reported cases. The WHO database contains nearly two million case reports of suspected ADRs for anonymous patients. Each report includes administrative data, patient data, ADR data, medication data, and other information. The database has 49 fields, though not all are filled in each report. The BCPNN is a feed-forward neural network that uses Bayes' law for learning and inference. It is self-organizing, suitable for parallel computers, and provides an efficient computational model. The BCPNN can handle missing data and is transparent, allowing for easy validation and checking. It is also efficient in training and searching through the database using a sparse matrix method. The Bayesian approach to signal generation involves calculating the probability of an ADR occurring with a specific drug. The BCPNN uses Bayes' law to calculate the probability of an ADR given a drug, and the probability of a drug given an ADR. This allows for the identification of drug-ADR combinations that are more likely to be associated than expected. The BCPNN is a powerful tool for detecting ADR signals from large datasets.