A residual feed intake (RFI) equation (ADFI = a + b₁ x BW^0.75 + b₂ x ADG), used to estimate ADFI in individual animals, was evaluated to determine if specification was correct and if predicted ADFI was equal to observed ADFI (validity). We evaluated 2 data sets of growing beef cattle using ordinary least squares (OLS), bootstrapping (BS) or a first order (FO) model in which final BW (BW_f) was a function of initial BW (BW₀) and cumulative DMI (DMI_c). The FO equation was BW_f = BW₀e^kDMIc. The RFI model was improperly specified (P < 0.10), predicted ADFI, both OLS and BS, differed from observed (P < 0.05), OLS and BS parameter estimates within and among data sets were not unique (P < 0.05) and the model failed all tests of external validity (P < 0.05). Functional forms of the variables were inconsistent with published classical energetic relationships among ADFI, BW and ADG. The intercept term, a, was either 0, negative or positive. The latter are inconsistent with biology; when BW and ADG are 0, a must likewise be 0. Parameter estimate b₁ indicated that ME intake at NE_g = 0 (ME_m) was either 0.280 x BW^0.75 or 0.165 x BW^0.75; both of these are greater (P < 0.05) than the NAS/NRC estimate of 0.131 x BW^0.75. Parameter estimate b₂ indicated that efficiency of ME utilization for gain was either outside or within the range of biological possibility. The FO model was internally and externally invalid for both data sets (P < 0.05) although R² were > 0.90. It appears that initial and final BW, as state variables, improve model fit in describing the relationship among ADFI, BW₀ and BW_f. Improper specification for the RFI model, parametric instability, lack of internal and external validity, irreproducibility of estimators between data sets and among published studies, and the random nature of residuals indicate the linear RFI model is a poor predictor of ADFI and of animal efficiency, for the data sets examined. A first order function (nonlinear) better describes relationships between BW change and ADFI. However, regardless of model, BW change and ADFI may be inadequate descriptors of efficiency.