We study the formation of ESG scores and rankings. In particular, we investigate the impact of aggregation rules when combining information on firms across categories, notably the E, S and G categories, into single ESG scores. Usual aggregation rules may bias scores toward the most dispersed category. We suggest a correction for this dispersion bias. We apply this correction to scores provided by two of the main score providers: Refinitiv and Bloomberg. We also provide simulation evidence. We show that the cross-dispersion bias may have a significant impact on ESG scores formation and that our proposed adjustment tends to weather it.