The therapeutic effects of neurofeedback (NFB) remain controversial. Here we show that NFB latency affects the ability of subjects to increase their parietal electroencephalographic alpha activity. Visual NFB was delivered either as soon as EEG was processed, or with a 250 or 500-ms added delay. After the training, a sustained increase in alpha average magnitude was observed only for the shortest latency tested. During the NFB sessions, learning profiles varied with latency, as shown with the adaptive Neyman test. The steepest growth of alpha magnitude was observed also for the shortest latency. For all latencies, NFB affected only the alpha spindle occurrence rate, not their duration or amplitude. The effectiveness of NFB only at the short-latency could explain the failures of some of the previous studies, where this parameter was neither controlled nor documented. Clinical practitioners and manufacturers of NFB equipment should include latency in their specifications, and enable short-latency operations.