This study aimed to evaluate the adequacy of seven non-linear models (France, Orskov and McDonald, Gompertz, exponential, logistic, two-pool exponential and two-pool logistic) in the adjustment of the curve and in the generation of parameters of cumulative gas production from five tropical feedstuffs (rice hulls, sugarcane, cassava chips, turnip by-product, and peach-palm by-product) used in ruminant nutrition. To this end, the feedstuffs were incubated in vitro in graduated glass syringes together with a buffer inoculum solution, in triplicate. Gas production was read at 0, 2, 4, 6, 8, 10, 12, 24, 26, 28, 30, 32, 36, 48, 52, 54, 56, 60 and 72 h of incubation. The data were used to generate the parameters of each model using the SAS statistical package. After the parameters were generated, the gas volume values were obtained at the aforementioned times, for each model, and these were compared with the values observed at incubation by using Model Evaluation System (MES) software. For the comparison, regression parameters were tested using Mayer’s test, in addition to the evaluation of the mean bias (MS), concordance correlation coefficient (CCC), and mean squared prediction error (MSPE). The France, Logistic, and Gompertz models, for rice hulls, and the Orskov and McDonald model, for cassava chips, were significant (P < 0.1) according to Mayer’s test, indicating lack of fit of the model. Besides presenting the lowest MSPE, the models that showed fit according to Mayer’s test were the two-pool logistic for rice hulls and cassava chips, and the two-pool exponential for sugarcane, turnip, and peach palm. Thus, the non-linear two-pool models are the most efficient in adjusting to the curve and in the generation of parameters of cumulative production of gases from the tested feedstuffs.