Objectives: The aim of our study was to assess the differences in facial emotional recognition (FER) between patients with first-episode psychosis (FEP), patients with multi-episode schizophrenia (SCH), and healthy controls (HC) and to find possible correlations of FER with psychopathology in the two patient groups. Methods: We performed a cross-sectional study enrolling 160 patients from two psychiatric hospitals in Croatia (80 FEP and 80 SCH) and 80 HC during the period from October 2015 until October 2017. Patients were assessed once during their hospital treatment, using the Penn Emotion Recognition Task for assessment of FER, rating scales for psychopathology and depression and self-reporting questionnaires for impulsiveness, aggression, and quality of life. Results: The number of correctly identified emotions significantly decreased from HC to FEP [Δ −7%; 95% confidence interval (CI) [−12% to −3%], effect size r = 0.30] and more markedly in SCH (Δ −15%; 95% CI [−25% to −10%], effect size r = 0.59) after the adjustment for age and gender and correction for multiple testing. Correct FER for negative emotions, but not for happiness and neutral emotions, had a statistically significant negative correlation with some features on the scales of psychopathology, impulsivity and aggression in both patient groups. Conclusions: Impairment of FER is present from the first episode of schizophrenia and increases further with multiple psychotic episodes, but it may depend on or contribute to clinical symptoms. Therefore, assessment of FER should be included in the clinical assessment and integrated in the plan of treatment from the beginning of the illness. (JINS, 2019, 25, 165–173)