This study was conducted to investigate the accuracy of predicting in vitro ruminal methane (CH4) production using volatile fatty acids (VFA) stoichiometric models \[CH4=0.5Ace-0.25Pro+0.5But-0.25Val\] (model 1), where CH4, Ace, Pro, But and Val are the production amounts of CH4, acetate, propionate, butyrate and valerate, respectively. Ten common feedstuffs, including four concentrates and six roughages with a wide range of chemical composition were incubated in serum bottles, and VFAs and CH4 production at 72 h were determined. The differences between the predicted and measured CH4 production were quantified using the model accuracy analysis. The results showed that the predicted CH4 production amounts were generally greater than the measured values obtained using the model 1, and the bias, slope and random error were 62.6%, 11.7% and 25.7%, respectively, indicating that fixed error exceeded 70%. By assuming 80% of total hydrogen being used for CH4 synthesis, the VFA stoichiometric model could be reexpressed as \[CH4=0.8(0.5Ace-0.25Pro+0.5But-0.25Val)\] (model 2). The root mean square prediction error (rMSPE=0.18) for model 2 was less than for model 1 (rMSPE =0.60). In addition, the bias, slope and random error of the model 2 were 2.1%, 5.7%, 92.3%, respectively, indicating that fixed error was less than 10%. In model 1, hydrogen formation resulting from VFA production were assumed to be totally consumed by methanogens for CH4 synthesis, without considering other pathways of hydrogen metabolism, which was the main factor resulting in the higher predicted values than the measured values.