Attention Bias Modification (ABM) aims to modulate attentional biases, but questions remain about its efficacy and there may be new variants yet to explore. The current study tested effects of a novel version of ABM, predictive ABM (predABM), using visually neutral cues predicting the locations of future threatening and neutral stimuli that had a chance of appearing after a delay. Such effects could also help understand anticipatory attentional biases measured using cued Visual Probe Tasks. 102 participants completed the experiment online. We tested whether training Towards Threat versus Away from Threat contingencies on the predABM would cause subsequent attentional biases towards versus away from threat versus neutral stimuli, respectively. Participants were randomly assigned and compared on attentional bias measured via a post-training Dot-Probe task. A significant difference was found between the attentional bias in the Towards Threat versus Away from Threat group. The training contingencies induced effects on bias in the expected direction, although the bias in each group separately did not reach significance. Stronger effects may require multiple training sessions. Nevertheless, the primary test confirmed the hypothesis, showing that the predABM is a potentially interesting variant of ABM. Theoretically, the results show that automatization may involve the process of selecting the outcome of a cognitive response, rather than a simple stimulus-response association. Training based on contingencies involving predicted stimuli affect subsequent attentional measures and could be of interest in future clinical studies.