The biological characteristics of trees in tropical dry savannas make it difficult to conduct inventories of tree density, and this has aroused interest in distance-based methods. This study proposes a distance based tree density estimator using Matern point processes, generating clustered spatial patterns. It was defined as the maximum likelihood estimator of the density, based on an approximate distribution of the distance from a random point to the pth nearest tree. It was compared with seven estimators identified in the literature as the most efficient. The estimators were compared on a benchmark of 10 point processes, with six being adjusted to observed tree patterns in six Mali savannas (West Africa). The proposed estimator was generally the most efficient. However, this result ignores that (i) all estimators do not require the same effort on the field, (ii) the point-processes benchmark was restricted to Matérn processes, and (iii) all estimators are not equivalent with respect to measurement errors.