Thirty various pelleted diets were given to broilers (8/diet) for in vivo measurements of dietary metabolisable energy (ME) value and digestibilities of proteins, lipids, starch and sugars from day 27 to day 31, with ad libitum feeding and total collection of excreta. Water excretion was also measured. Amino acid formulation of diets was done on the basis of ratios to crude proteins. Mean in vivo apparent ME values corrected to zero nitrogen retention (AMEn) were always lower than the AMEn values calculated for adult cockerels using predicting equations from literature based on the chemical analyses of diets. The difference between mean in vivo AMEn values and these calculated AMEn values increased linearly with increasing amount of wheat in diets (P = 0.0001). Mean digestibilities of proteins, lipids and starch were negatively related to wheat introduction (P = 0.0001). The correlations between mean in vivo AMEn values and diet analytical parameters were the highest with fibre-related parameters, such as water-insoluble cell-walls (WICW) (r = −0.91) or Real Applied Viscosity (RAV) (r = −0.77). Thirteen multiple regression equations relating mean in vivo AMEn values to dietary analytical data were calculated, with R2 values ranging from 0.859 to 0.966 (P = 0.0001). The highest R2 values were obtained when the RAV parameter was included in independent variables. The direct regression equations obtained with available components (proteins, lipids, starch, sucrose and oligosaccharides) and the indirect regression equations obtained with WICW and ash parameters showed similar R2 values. Direct or indirect theoretical equations predicting AMEn values were established using the overall mean in vivo digestibility values. The principle of indirect equations was based on the assumption that WICW and ashes act as diluters. Addition of RAV or wheat content in variables improved the accuracy of theoretical equations. Efficiencies of theoretical equations for predicting AMEn values were almost the same as those of multiple regression equations. Water excretion was expressed either as the water content of excreta (EWC), the ratio of water excretion to feed intake (WIR) or the residual value from the regression equation relating water excretion to feed intake (RWE). The best regression predicting EWC was based on sucrose, fermentable sugars (lactose + oligosaccharides) and chloride variables, with positive coefficients. The best equations predicting WIR or RWE contained the sugar and chloride variables, with positive coefficients. Other variables appearing in these equations were AMEn or starch with negative coefficients, WICW, ‘cell-wall-retained water’, RAV or potassium with positive coefficients.