Empirical research using the Empathic Accuracy Task (EAT) has suggested that schizophrenia patients and people with schizotypal personality disorder exhibit lower empathic accuracy than healthy people. However, empathic accuracy in a subclinical sample with high levels of schizotypy has seldom been studied. Our study aimed to investigate empathy in a subclinical sample using the Chinese version of the EAT and a self-report empathy measure. Forty participants with high levels of schizotypy (HS participants) and 40 with low levels of schizotypy (LS participants), as measured by the Schizotypal Personality Questionnaire (SPQ), were recruited. All participants completed the Chinese version of the EAT and the self-report Questionnaire of Cognitive and Affective Empathy. Empathic accuracy (EA) scores and the intra-individual variability of EA scores were calculated. Independent samples t tests and Pearson correlation analyses were performed to examine group differences in empathy and the relationship between empathy and schizotypy respectively. HS participants exhibited reduced EA for both positive and negative videos, and larger intra-individual variability of EA for negative videos than LS participants. However, HS and LS participants did not differ in self-report cognitive empathy. Moreover, the interpersonal dimension of the SPQ was negatively correlated with EAT performance and self-report cognitive empathy in LS participants. Individuals with HS show poorer performance-based EA but relatively intact self-report cognitive empathy. This study provides empirical evidence for the ontogeny of empathy deficits in subclinical populations at risk of developing schizophrenia, supporting early interventions for social cognitive deficits.