Combined with other CMB experiments, the WMAP survey provides an accurate estimate of the baryon density of the Universe. In the framework of the standard Big Bang Nucleosynthesis (BBN), such a baryon density leads to predictions for the primordial abundances of $^{4}$He and D in good agreement with observations. However, it also leads to a significant discrepancy between the predicted and observed primordial abundance of $^{7}$Li. Such a discrepancy is often termed as 'the lithium problem'. In this paper, we analyze this problem in the framework of scalar-tensor theories of gravity. It is shown that an expansion of the Universe slightly slower than in General Relativity before BBN, but faster during BBN, solves the lithium problem and leads to $^4$He and D primordial abundances consistent with the observational constraints. This kind of behavior is obtained in numerous scalar-tensor models, both with and without a self-interaction potential for the scalar field. In models with a self-interacting scalar field, the convergence towards General Relativity is ensured without any condition, thanks to an attraction mechanism which starts to work during the radiation-dominated epoch.