Background: The automatic tendency to attend to, positively evaluate and approach alcohol related stimuli has been found to play a causal role in problematic alcohol use and can be retrained by computerised Cognitive Bias Modification (CBM). In spite of CBMs potential as an internet intervention, little is known about the efficacy of web-based CBM. The study described in this protocol will test the effectiveness of web-based CBM in a double blind randomised controlled trial with a 2 (attention bias retraining: real versus placebo) x 2 (alcohol/no-go training: real versus placebo) x 2 (approach bias retraining: real versus placebo) factorial design. Methods/design. The effectiveness of 12 sessions of CBM will be examined among problem drinkers aged 18-65 who are randomly assigned to one of the eight CBM conditions, after completing two modules of a validated cognitive behavioural intervention, DrinkingLess. The primary outcome measure is the change in alcohol use. It is expected that, for each of the CBM interventions, participants in the real CBM conditions will show a greater decrease in alcohol use than participants in the placebo conditions. Secondary outcome measures include the percentage of participants drinking within the limits for sensible drinking. Possible mediating (change in automatic biases) and moderating (working memory, inhibition) factors will be examined, as will the comparative cost-effectiveness of the various CBM strategies. Discussion. This study will be the first to test the relative efficacy of various web-based CBM strategies in problem drinkers. If proven effective, CBM could be implemented as a low-cost, low-threshold adjuvant to CBT-based online interventions for problem drinkers. Trial registration. Netherlands Trial register: NTR3875. © 2013 van Deursen et al.; licensee BioMed Central Ltd.