OBJECTIVES: Several studies have demonstrated that Risperidone Long-Acting Injectable (RLAI) reduces psychotic relapses, hospitalization and resources used among schizophrenic patients with poor adherence to oral medication.  However, the magnitude of such reductions depends upon the baseline relapse rate.  This study is aimed at identifying the thresholds of relapse risk at which RLAI is cost-effective, compared to Oral Olanzapine (OO), Oral Quetiapine (OQ) and Haloperidol Decanoate (HD). METHODS: A Markov model was developed to simulate the natural history of schizophrenia for patients who have poor adherence and high risk of relapse with oral medication.  The strategies compared were starting treatment with RLAI, OO, OQ or HD.  Relapse probabilities, adherence levels, side effects and treatment switching were derived from long-term observational data.  Resource use and costs were obtained from Mexican public institutions.  Patients transit through different health states in the model on a monthly basis over a 10-year time horizon. Incremental cost and effectiveness outcomes were discounted at 3% annually. RESULTS: In patients with Baseline Annual Relapse Rate (BARR) equal or greater than 72.5%, RLAI is the most effective and less costly treatment (cost-saving) if effectiveness is measured in terms of QALYs or relapses averted.  In this case, RLAI produces 0.27 (5.13-5.4) additional discounted QALYs and avoids 3.97 relapses, compared to OO. According to the WHO Commission on Macroeconomics and Health, RLAI is highly cost-effective (≤1xGDP per-capita/QALY gained) in patients with a BARR between 63% and 72.5% and cost-effective (≤3xGDP per-capita /QALY gained) in patients with a BARR between 35% and 63%. CONCLUSIONS: In all published naturalistic studies comparing RLAI with oral medication, where the selection of patients to receive RLAI is left to the physicians, the BARR is greater than 0.71, which suggest that using RLAI in Mexico with similar criteria of patient selection would result in a cost-saving strategy.