Objective: To test serum tau as a predictor of neurological outcome after cardiac arrest. Methods: We measured the neuronal protein tau in serum at 24, 48, and 72 hours after cardiac arrest in 689 patients in the prospective international Target Temperature Management trial. The main outcome was poor neurological outcome, defined as Cerebral Performance Categories 3–5 at 6 months. Results: Increased tau was associated with poor outcome at 6 months after cardiac arrest (median = 38.5, interquartile range [IQR] = 5.7–245ng/l in poor vs median = 1.5, IQR = 0.7–2.4ng/l in good outcome, for tau at 72 hours, p < 0.0001). Tau improved prediction of poor outcome compared to using clinical information (p < 0.0001). Tau cutoffs had low false-positive rates (FPRs) for good outcome while retaining high sensitivity for poor outcome. For example, tau at 72 hours had FPR = 2% (95% CI = 1–4%) with sensitivity = 66% (95% CI = 61–70%). Tau had higher accuracy than serum neuron-specific enolase (NSE; the area under the receiver operating characteristic curve was 0.91 for tau vs 0.86 for NSE at 72 hours, p = 0.00024). During follow-up (up to 956 days), tau was significantly associated with overall survival. The accuracy in predicting outcome by serum tau was equally high for patients randomized to 33 °C and 36 °C targeted temperature after cardiac arrest. Interpretation: Serum tau is a promising novel biomarker for prediction of neurological outcome in patients with cardiac arrest. It may be significantly better than serum NSE, which is recommended in guidelines and currently used in clinical practice in several countries to predict outcome after cardiac arrest. Ann Neurol 2017;82:665–675.