Attending to an object can strongly modulate the neural processing of this object. Using average EEG, it was shown that small differences in the focus of attention yield large and long lasting changes in brain dynamics (Plomp et al, 2009). Here, we show that these changes can successfully be decoded on the single-trial level by combining Fisher linear discriminant (FLD) analysis with principal component analysis (PCA). We presented two expanding streams of lines originating at one central element. In each trial, a cue indicated to which stream observers had to attend. The last elements of the two streams were only half a degree of visual angle apart from each other, so that the attention shift was within a quarter degree. Still, we could decode the focus of attention (left or right stream attended) with an average precision of 72% (ranging from 57% to 84% depending on observer). For classifiers based on single time points of the data, highest precision was obtained between 100 and 150 ms after stimulus onset. Because the stimulus was identical in both cueing conditions, we show that it is possible to decode internally generated attention signals from single-trial EEG.