Abundance of small populations of large mammals may be assessed using complete counts of the different individuals detected over a time period, so-called minimum detected size (MDS). However, as population is growing larger and its distribution is expanding wider, the risk of under-estimating population size using MDS is increasing sharply due to the rarely fulfilled assumption of perfect detection of all individuals of the population, and as a result, the need to report uncertainty in population size estimates becomes crucial. We addressed these issues within the framework of the monitoring of the critically endangered Pyrenean brown bear population that was on the edge of extinction in the mid-1990s with only five individuals remaining, but was reinforced by 11 bears originated from Slovenia since then. We used Pollock's closed robust design (PCRD) capture-recapture models applied to the cross-border non-invasive sampling data from France, Spain and Andorra to provide the first published annual abundance estimates of the Pyrenean brown bear population and its trends over time. Annual population size increased and displayed a fivefold rise between 2008 and 2020, reaching > 60 individuals in 2020. Detection heterogeneity among individuals may stem from intraspecific home range size disparities making it more likely to find signs of individuals who move more. We found a lower survival rate in cubs than in adults and subadults, since the formers suffer from more mortality risks (such as infanticides, predations, mother death or abandonments) than the latters. Our study provides evidence that the PCRD capture-recapture modelling approach can provide reliable estimates of the size of and trend in large mammal populations, while minimizing bias due to inter-individual heterogeneity in detection probabilities and allowing the quantification of sampling uncertainty surrounding these estimates. Such information is vital for informing management decision-making and assessing population conservation status.