The random error characteristics of optical voltage sensors (OVS) directly affect the measurement accuracy. Sequence variance is a commonly used method to evaluate the characteristics of the OVS random error. However, sequence variance does not model the OVS noise itself and does not identify the noise coefficient in terms of OVS noise signature analysis. Based on sequence variance method, the improved sequential analysis of variance method is proposed to analyze the random error characteristics of OVS. After the OVS test data is processed by the improved sequential analysis of variance method, the least square fitting is performed, and the coefficients of each random error term are accurately solved. With the coefficients, the influencing factors are identified, and the corresponding error suppression measures are proposed. The experiments results demonstrate that the improved sequential analysis of variance method is different from the sequence variance method, the sequence variance method fluctuates and gradually increases when the correlation time is more than half of the total measurement duration, while the improved sequential analysis of variance method always remains relatively stable.