We describe and demonstrate the use of quantile regression for modeling errors associated with streamflow predictions. Quantile regression has been used rarely in hydrology, yet its insensitivity to outliers and avoidance of parametric assumptions makes it suitable for representing quantiles of datasets (such as streamflow) that often have skewed or otherwise irregular distributions. For illustration, a local linear quantile regression framework is used to bias-correct and estimate error bounds for deterministic hourly flood predictions, and also to calibrate ensemble seasonal runoff predictions, thereby reducing their bias and improving forecast reliability.