Uncertainties in tumor position during intensity-modulated radiotherapy (IMRT) plan optimization are usually accounted for by adding margins to a clinical target volume (CTV), or additionally, to organs at risk (OAR). The former approach usually favors target coverage over OAR protection, whereas the latter does not account for correlation in target and OAR movement. We investigate a new approach to incorporate systematic errors in tumor and organ position. The method models a distribution of systematic errors due to setup error and organ motion with displaced replicas of volumes of interest, each representing the patient geometry for a possible systematic error, and maximizes a score function that counts the number of replicas meeting dose or biological constraints for both CTV and OAR. Dose constraints are implemented by logistic functions of Niemierko's generalized model of equivalent uniform dose (EUD). The method is applied to prostate and nasopharynx IMRT plans, in which CTV and OAR each consists of five replicas, one representing no error (the position in the planning CT) and the other four discrete systematic setup displacements in one dimension with equal probability. The resulting IMRT plans are compared with those from two other EUD-based optimizations: a standard planning target volume (PTV) approach consisting of a single replica of each OAR in the planned position and a single PTV encompassing all CTV replicas, and a PTV-PRV approach consisting of a single PTV and a single planning risk volume (PRV) for each OAR encompassing all replicas. When systematic error is present, multiple-replica optimization provides better critical organ protection while maintaining similar target coverage compared with the PTV approach, and provides better CTV-to-OAR therapeutic ratio compared with the PTV-PRV instances where there is substantial PTV-PRV overlap. The method can be used for other systematic errors due to organ motion and deformation.