Growing pigs are often fed below ad libitum to increase their feed efficiency and carcass leanness. When energy supply is under control, precision feeding is implemented through the amino acids (AA). As the AA requirement depends on the body weight (BW) for the maintenance part and on its daily variation (ΔBW) for the growth part, the adequacy between requirements and supplies on day D+1 depends on the adequacy of predicted BWD+1 and ΔBWD+1. Data sets from four trials were used to forecast BW from time series analyses based either on multivariate adaptive regression splines (MARS) or double exponential smoothing (HWα) methods using the k latest data (8, 14 or 20). Pigs (n = 117) were group-housed and restrictively fed, and their BW was recorded daily and individually with an automatic scale (n = 11 736). With HW0.6, the RMSEP of BWD+1 was the smallest one (1.21 kg) and not influenced by k. Linear regression on the l latest forecasted BW was used to assess ΔBWD+1. At the beginning of the trial, ΔBWD+1 was more difficult to predict from BW forecasted with MARS than with HW0.6. Descriptive statistics of individual variation of ΔBWD+1 based on MARS and HW0.6 were comparable with k = l = 20 only after removal of the first 19 days. Compared to other methods studied, the method HW0.6 seems to be the best compromise to forecast BWD+1 and ΔBWD+1 of restrictively fed pigs.