Subjective tinnitus, the perception of sound in the absence of any sound source, is routinely assessed using questionnaires. The subjective nature of these tools hampers objective evaluation of tinnitus presence, severity and treatment effects. Late auditory evoked potentials (LAEPs) might be considered as a potential biomarker for assessing tinnitus complaints. Using a multivariate meta-analytic model including data from twenty-one studies, we determined the LAEP components differing systematically between tinnitus patients and controls. Results from this model indicate that amplitude of the P300 component is lower in tinnitus patients (standardized mean difference (SMD) = -0.83, p < 0.01), while latency of this component is abnormally prolonged in this population (SMD = 0.97, p < 0.01). No other investigated LAEP components were found to differ between tinnitus and non-tinnitus subjects. Additional sensitivity analyses regarding differences in experimental conditions confirmed the robustness of these results. Differences in age and hearing levels between the two experimental groups might have a considerable impact on LAEP outcomes and should be carefully considered in future studies. Although we established consistent differences in the P300 component between tinnitus patients and controls, we could not identify any evidence that this component might covary with tinnitus severity. We conclude that out of several commonly assessed LAEP components, only the P300 can be considered as a potential biomarker for subjective tinnitus, although more research is needed to determine its relationship with subjective tinnitus measures. Future trials investigating experimental tinnitus therapies should consider including P300 measurements in the evaluation of treatment effect.