The collection, characterization and conservation of variability within a species remain the key for variety improvement. In Cameroon, maize production is limited by stress conditions common to tropical countries, including drought, pests, diseases, acid soils, etc. One of the impediments to the progress in selection targeted in solving these constraints is the lack of variability in the germplasm used as base material in breeding programs. In these programs, little emphasis has been placed on the study and use of the variability available within maize landraces. In year 2004 Ninety-height maize accessions (eighty-six landraces and twelve improved varieties) were collected from the lowland humid forest zone of Cameroon and evaluated on the basis of fifteen agro-morphological characters in a randomized complete block design with three replications. The objectives of this study were to evaluate individual and group variation within lowland humid forest zone maize landraces, to identify the most important characters to discriminate and classify these accessions, and to identify outstanding accessions with desirable traits to be used as gene donor in breeding programs. The analysis of variance detected highly significant differences among the accessions for all characters studied. The range of variation was large for most characters, indicating the existence of variability among accessions. A multivariate analysis was carried out and principal component analysis identified six principal components (PC), which together explained 81% of the variation. Phenological characteristics (number of days from planting to 50% silking and number of days from planting to 50% tasseling) and architectural characteristic (Plant height, ear height, number of leaves above the upper ear) were the best descriptors as revealed by the first PC (26% of the variation). The second PC (23% of the variation) was associated with ear and kernel characteristics (ear length, ear shape, 1 000 grain weight, and yield) and DTS. Cluster analysis allowed the classification of accessions into five groups; each being a pool where specific characters can be found. There was no clear cut for the assignment of accession 86 and 29 to specific clusters. On the biplot constructed from the first and second PCs, overlapping of accessions 96 and 100 (two three-way hybrids with two parents in common), was probably an expression of their similarity. This was also the case with accessions 41 and 75 (two landraces). From the plant texture and yield potential expressed by some accessions, it could be assumed that they were local varieties which have been subjected to intentional or unintentional inflow of foreign improved germplasm.