This study investigated the effect of different feeding regimes on the cobalt sorption capacity of anaerobic granular sludge from a full-scale bioreactor treating paper mill wastewater. Adsorption experiments were done with non-fed granules in monometal (only Co) and competitive conditions (Co and Ni in equimolar concentrations). In order to modify the extracellular polymeric substances and sulfides content of the granules, the sludge was fed for 30 days with glucose (pH 7, 30 °C, organic loading rate = 1.2 g glucose l⁻¹ day⁻¹) in the presence (COD/SO₄²⁻ = 1) or absence of sulfate. The partitioning of the sorbed cobalt between the exchangeable, carbonates, organic matter/sulfides and residual fractions was determined using a sequential extraction procedure (modified Tessier). Experimental equilibrium sorption data for cobalt were analysed by the Langmuir, Freundlich and Redlich–Peterson isotherm equations. The total Langmuir maximal sorption capacity of the sludge fed with glucose and sulfate loaded with cobalt alone displayed a significantly higher maximal cobalt sorption (Q_max = 18.76 mg g⁻¹ TSS) than the sludge fed with glucose alone (Q_max = 13.21 mg g⁻¹ TSS), essentially due to an increased sorption capacity of the exchangeable (30–107%) and organic/sulfides fractions (70–30%). Environmental scanning electron microscopy coupled with an energy dispersive X-ray analysis of granular cross-sections showed that mainly iron minerals (i.e. iron sulfides) were involved in the cobalt accumulation. Moreover, the sorbed cobalt was mainly located at the edge of the granules. The sorption characteristics of the exchangeable and carbonates fractions fitted well to the Redlich–Peterson model (intermediate multi-layer sorption behaviour), whereas the sorption characteristics of the organic matter/sulfides and residual fractions fitted well to the Langmuir model (monolayer sorption behaviour). The organic matter/sulfides fraction displayed the highest affinity for cobalt for the three sludge types investigated.