The N2pc component is the most-used MEG/EEG signal for tracking feature-based target selection. However, the spatial resolution of the N2pc is limited, because it is based on activity differences between hemispheres. Here we demonstrate that multivariate analyses of raw EEG data provide a much more fine-grained spatial profile of feature-based target selection. Experiment 1 used a visual search task in which colour-defined targets appeared on the horizontal or vertical midline. In Experiment 2, search displays contained a target that appeared at one of eight possible locations among seven distractors. Using pattern classification, we show that information from EEG signals can be used to decode target positions on the vertical midline (above versus below fixation) as well as within the same visual field quadrant, which cannot be achieved using standard N2pc methodology. Classification accuracy increased rapidly from about 200 ms after display onset. To further characterize the spatial precision of this signal, we used a forward encoding model to construct a cortical tuning function that describes the relationship between target position and multivariate EEG. This model is fully invertible, allowing us to construct hypothetical topographic activation maps for targets on the horizontal/vertical midline that were never actually shown during Experiment 2, and that were then tested against the horizontal/vertical position data of Experiment 1. The constructed maps were statistically indistinguishable from the real pattern of neural activity, providing independent validation. Our findings demonstrate the power of multivariate EEG analysis to track feature-based target selection with high spatial and temporal precision.