The concept of the biological activity spectrum was introduced to describe the properties of biologically active substances. The PASS software predicts over 300 pharmacological effects and biochemical mechanisms based on the structural formula of a substance. It can be used to find new targets for ligands or new ligands for biological targets. A WWW interface for PASS has been developed, allowing online prediction of biological activity spectra. Bioinformatics is moving from sequence analysis to finding new drug targets and ligands. Ligands can bind to similar proteins, and structure-activity relationships can help find new ligands. The biological activity spectrum is a list of activity names reflecting a substance's interaction with biological entities. The goal is to provide full information on a substance's biological activities. Structure descriptors called 'multilevel neighbourhoods of atoms' (MNAs) are used for chemical structure representation. These descriptors are generated recursively and include atom valencies and partial charges. The prediction algorithm starts by generating MNAs for a substance. Probability estimates for each activity are calculated using a mathematical formula. The prediction result is a table with activity names and probability values. The accuracy of predictions varies, with an average of 89% (leave-one-out cross-validation). The Internet version of the program, PASS Inet, includes 31,000 biologically active substances and predicts 319 types of pharmacological effects. Users can submit a Molfile or draw a structure to obtain the biological activity spectrum. The results include the number of chemical descriptors and comments on the interpretation. If Pa > 0.7, the substance is likely to exhibit the activity. If 0.5 < Pa < 0.7, the activity is likely but less probable. If Pa < 0.5, the activity is unlikely, but confirmation may indicate a new chemical entity. Figure 1 shows the predicted activity spectrum for nipradilol, which has 50 descriptors and 26 activities. Nipradilol is both a beta-1 adrenoreceptor antagonist and agonist, indicating interaction with the receptor but unclear mechanism.The concept of the biological activity spectrum was introduced to describe the properties of biologically active substances. The PASS software predicts over 300 pharmacological effects and biochemical mechanisms based on the structural formula of a substance. It can be used to find new targets for ligands or new ligands for biological targets. A WWW interface for PASS has been developed, allowing online prediction of biological activity spectra. Bioinformatics is moving from sequence analysis to finding new drug targets and ligands. Ligands can bind to similar proteins, and structure-activity relationships can help find new ligands. The biological activity spectrum is a list of activity names reflecting a substance's interaction with biological entities. The goal is to provide full information on a substance's biological activities. Structure descriptors called 'multilevel neighbourhoods of atoms' (MNAs) are used for chemical structure representation. These descriptors are generated recursively and include atom valencies and partial charges. The prediction algorithm starts by generating MNAs for a substance. Probability estimates for each activity are calculated using a mathematical formula. The prediction result is a table with activity names and probability values. The accuracy of predictions varies, with an average of 89% (leave-one-out cross-validation). The Internet version of the program, PASS Inet, includes 31,000 biologically active substances and predicts 319 types of pharmacological effects. Users can submit a Molfile or draw a structure to obtain the biological activity spectrum. The results include the number of chemical descriptors and comments on the interpretation. If Pa > 0.7, the substance is likely to exhibit the activity. If 0.5 < Pa < 0.7, the activity is likely but less probable. If Pa < 0.5, the activity is unlikely, but confirmation may indicate a new chemical entity. Figure 1 shows the predicted activity spectrum for nipradilol, which has 50 descriptors and 26 activities. Nipradilol is both a beta-1 adrenoreceptor antagonist and agonist, indicating interaction with the receptor but unclear mechanism.