Background: Most neglect treatment studies focus on automatic re-orientation procedures, assuming a deficit in automatic processes. We compare an automatic- and a controlled procedure, using the endogenous and exogenous variants of Posner's covert shift of attention task. Method: In two experiments, neglect patients and patients with a right hemispherical stroke without neglect performed three blocks of Posner's covert shift of attention task (Posner Task) on two days. In Study 1 we used endogenous cues, in Study 2, exogenous cues. Results: In the endogenous task, neglect patients improved significantly with valid left-sided cues between block 1 and 2 on Day 1, subsequently showing a plateauing. They also showed a gradual improvement on invalid trials on both days. In the exogenous condition, all participants responded only increasingly faster on trials with a long stimulus onset asynchrony. Practicing on both tasks led to fewer omissions for left-sided targets, minimally in the exogenous and clearly in the endogenous condition. Conclusion: In line with prior neuroanatomical studies, our study shows that practicing an endogenous, but not an exogenous, visuospatial attention task leads to significant improvements in neglect patients, especially for invalid trials, suggesting that neglect treatments based on top-down strategies should be given more attention.