Although the brown bear (Ursus arctos) population in Abruzzo (central Apennines, Italy) suffered high mortality during the past 30 years and is potentially at high risk of extinction, no formal estimate of its abundance has been attempted. In 2004, the Italian Forest Service and Abruzzo National Park applied DNA-based techniques to hair-snag samples from the Apennine bear population. Even though sampling and theoretical limitations prevented estimating population size from being the objective of these first applications, we extracted the most we could out of the 2004 data to produce the first estimate of population size. To overcome the limitations of the sampling strategies (systematic grid, opportunistic sampling at buckthorn [Rhamnus alpina] patches, incidental sampling during other field activities), we used a multiple data-source approach and Huggins closed models implemented in program MARK. To account for model uncertainty, we averaged plausible models using Akaike weights and estimated an unconditional population size of 43 bears (95% CI  =  35–67). We urge caution in interpreting these results because other expected but undefined sources of heterogeneity (i.e., gender) may have biased this estimate. The low capture probability obtained through the systematic grid prevented the use of this sampling technique as a stand-alone tool to estimate the Apennine bear population size. Therefore, further applications in this direction will require a substantial improvement of field procedures, the use of a multiple data-source approach, or both. In this perspective, we used Monte Carlo simulations to compare the relative performance of the 3 sampling approaches and discuss their feasibility to overcome the problem of small and sparse DNA data that often prevent reliable capture–mark–recapture applications in small bear populations.