The promise of fMRI to detect robust inter-individual differences in health and behavioural outcomes and its potential use as a biomarker has been increasingly called into question. In a recent study (Hoeppli et al., 2022), the authors argued that, in the context of pain perception, fMRI was sensitive to small differences in applied noxious stimuli within an individual, but failed at predicting inter-individual differences. Here, we test in six studies whether this absent predictive power can be restored by accounting for inter-individual variance and the use of clearly distinguishable experimental conditions. Out of six studies, only one showed a statistically significant relationship between absolute NPS scores (a multivariate pain classifier based on fMRI) and pain ratings (r = .44, p < .01). In all other studies we replicated the null-finding by Hoeppli et al. (2022). However, we observed a significant positive linear relationship between the difference in NPS and the difference in pain ratings between two painful stimuli across all analysed studies (pooled result: r = .29, p < .01). More importantly, all analysed studies showed a significant positive linear relationship between the difference in NPS (comparing a painful and a non-painful, yet warm condition) and the rating of the painful condition (pooled result: r = .44, p < .01). Our results demonstrate that a fMRI-based multivariate pain classifier has predictive power for inter-individual differences in pain ratings. We conclude that two main factors influence its success: Inter-individual variance in the fMRI signal and the choice of experimental conditions.