We report psychophysical measurements of the temporal resolution of visual feature binding. A feature pair, such as "red and leftward tilt", was temporally alternated with another, such as "green and rightward tilt". When the features are spatially superimposed, observers can report the pairing for stimulus alternation rates less than 15 Hz. But when features are spatially separated, thresholds fall to less than 3 Hz. Attaching the spatially separated features to a common object or surface has little effect on the temporal thresholds. Changing the distance between the spatially separated features also has little effect. These results suggest that the binding of spatially separated features and the binding of spatially superimposed features are mediated by distinct mechanisms. The high temporal resolution in the spatially superimposed case may reflect cortical neurons with localized receptive fields selective to multiple features, whereas reporting spatially separated features likely requires an attention-mediated process.