This paper presents results of a study measuring user performance in a procedural task using Spatial Augmented Reality (SAR). The task required participants to press sequences of buttons on two control panel designs in the correct order. Instructions for the task were shown either on a computer monitor, or projected directly onto the control panels. This work was motivated by discrepancies between the expectations from AR proponents and experimental findings. AR is often promoted as a way of improving user performance and understanding. With notable exceptions however, experimental results do not confirm these expectations. Reasons cited for results include limitations of current display technologies and misregistration caused by tracking and calibration errors. Our experiment utilizes SAR to remove these effects. Our results show that augmented annotations lead to significantly faster task completion speed, fewer errors, and reduced head movement, when compared to monitor based instructions. Subjectively, our results show augmented annotations are preferred by users.