This research aims to study the psychometric properties and theoretical structure of the Serbian version of the Mathematics Teaching Efficacy Beliefs Instrument (MTEBI) which was developed by Enochs, Smith and Huinker (2000). The sample consisted of 252 preservice primary and preschool teachers from teacher education faculties in Serbia. The original MTEBI indicated acceptable reliability (α=0.779). The results of confirmatory factor analysis indicate that the fit of the original MTEBI model to the data is not acceptable, but the re-specified model (MTEBI12) shows good fit and acceptable reliability (α=0.742). These findings indicate that a short 12 item version of MTEBI possesses adequate psychometric properties and is applicable to the sample of respondents in the Republic of Serbia.