The objectives of this study were to estimate the stability of dry biomass production of elephant grass genotypes under an annual harvest regime, in soil-climatic conditions of the Northern region of the Rio de Janeiro State, Brazil, and to compare methodologies for stability analyses of Yates and Cochran (YC), Plaisted and Peterson (PP), Annicchiarico (ANN), Lin and Binns (LB), Huenh (HU), and Kang and Phan (KP). A randomized block design with 83 treatments and two replicates was adopted. Four annual harvests were performed (2012-2015) and dry matter yield (DMY, t.ha-1.year-1) was evaluated. Individual and combined analyses of variance for DMY revealed significant effects for genotypes, harvests, and for the genotype x harvest interaction at the probability levels of 1 and 5%. Genotypes indicated by the YC and PP methods are associated with higher stability and lower DMY. The weighting of KP with YC and PP was highly effective in associating stability with DMY. The LB and ANN methods showed strong agreement with each other and produced similar classifications as to phenotypic stability, and so we recommend using one or the other. Genotypes Elefante Cachoeiro do Itapemirim, Cuba-116, Taiwan A-46, P241 Piracicaba, Taiwan A-144, Cameroon - Piracicaba, 10 AD IRI, Guaçu/I,Z,2, Mineirão IPEACO, Taiwan A-121, IJ7125 cv EMPASC308, 903-77, Mole de Volta Grande, and Porto Rico 534-B showed high stability and DMY, standing out as promising genotypes for the soil-climatic conditions of the Northern region of the Rio de Janeiro State. The methodologies based on ANOVA and non-parametric analyses were complementary and increased reliability in the recommendation of genotypes.