Non-metric multidimensional scaling (NMDS) is one of the most used ordination methods in vegetation studies. Its application in statistical software often requires the choice of several options on which the accuracy of results may depend. Previous studies evidenced how similarity indexes and data standardization affect NMDS efficiency. Yet they did not focus on the case of sample size that could be of great importance since sample size highly impact estimation efficiency in forest studies. The here-reported study focused on the combined effect of sample size, similarity/dissimilarity indexes, data standardization and type of data matrix (abundance and binary) on NMDS efficiency based on real data from the Lama Forest Reserve, a dense semi-deciduous forest in Southern-Benin. NMDS ordination efficiency was assessed with two criteria: the Spearman rank correlation and the s-stress value. Results revealed that all the four factors influenced the efficiency of the NMDS. Binary matrices gave better results than abundance matrices from which they derived. However, this could not justify the simplification of an abundance matrix into a binary matrix since the risk of ecological information loss is tremendous. Among standardizations, samples (plots) standardization to equal totals (i.e. assigning sample the same weight) gave the best results. Similarity/dissimilarity indexes of Jaccard and Sorensen performed equally whatever the nature of the matrix. However, with binary matrices, similarity index of Sokal & Michener performed better. The sample size significantly affected mean values of criteria of efficiency as well as dispersion around them. Thus, a quadratic relationship was noted between s-stress and sample size. The optimal sample size was lower (75 plots) for the binary matrices than for the abundance ones (90 plots).