Context: In most natural populations, exhaustive counts are not possible and estimates need to be derived from partial sampling by using analytical methods that account for biological processes, sampling errors and detection probability. The methods available have contrasting pitfalls and payoffs in relation to the assumptions made but are seldom contrasted on the same population.

Aims: We compared density estimates derived by different sampling methods. Despite the real density being unknown, the ‘soft’ validation of density estimates might help to better understand the possible pitfalls and payoffs of each method. This was done in three closed populations and with three different habitat typologies to disentangle the effects of different capture-detection processes to those introduced by the method itself.

Methods: We considered the problem of estimating population density of the endemic Balearic lizard, Podarcis lilfordi, in three island populations. We compared estimates derived by distance sampling (LT) in three types of habitat with those calculated from a simultaneous 3-day capture–mark–recapture study. Capture histories of marked individuals were used to estimate density using spatially explicit capture–recapture models (SECR) and a capture–mark–recapture model without spatial data (CMR). Moreover, we empirically assessed the influence of survey duration by extending the survey in the largest island to five occasions. The real population density was unknown and absolute accuracy of each method cannot be assessed; nevertheless, relative estimates might be informative.

Key results: LT estimates had the greatest coefficient of variation in vegetated habitats, corresponding to possible departures from model assumptions. SECR estimates differed among islands and were from 12% to 37% lower than those derived by LT but only in the largest islands with high and dense vegetation. CMR estimates depended on the number of occasions whereas SECR did not and showed lower variance. LT and SECR estimates showed differences across islets.

Conclusions: Line-transect and capture–recapture methods gave comparable results but the interaction between recapture processes and habitat types should be considered when inferring density to the whole area. We found density estimates between 1500 and 2500 individuals ha^–1, being a higher value than those found for lizards in continental regions.

Implications: Pitfalls and payoffs of each method are discussed to optimise experimental design in estimating population density.