Broad-scale fire regime modelling is frequently based on large ecological and (or) administrative units. However, these units may not capture spatial heterogeneity in fire regimes and may thus lead to spatially inaccurate estimates of future fire activity. In this study, we defined homogeneous fire regime (HFR) zones for Canada based on annual area burned (AAB) and fire occurrence (FireOcc), and we used them to model future (2011-2040, 2041-2070, and 2071-2100) fire activity using multivariate adaptive regression splines (MARS). We identified a total of 16 HFR zones explaining 47.7% of the heterogeneity in AAB and FireOcc for the 1959-1999 period. MARS models based on HFR zones projected a 3.7-fold increase in AAB and a 3.0-fold increase in FireOcc by 2100 when compared with 1961-1990, with great interzone heterogeneity. The greatest increases would occur in zones located in central and northwestern Canada. Much of the increase in AAB would result from a sharp increase in fire activity during July and August. Ecozone-and HFR-based models projected relatively similar nationwide FireOcc and AAB. However, very high spatial discrepancies were noted between zonations over extensive areas. The proposed HFR zonation should help providing more spatially accurate estimates of future ecological patterns largely driven by fire in the boreal forest such as biodiversity patterns, energy flows, and carbon storage than those obtained from large-scale multipurpose classification units.