Background Body composition assessment in children with cerebral palsy (CP) is a challenge, specially the fat percentage. There are different methods that can be used to estimate the fat percentage in this population, such as anthropometric equations, but there is still a need to determine which is the best and most accurate. The purpose of the study was to determine the method that best estimates the fat percentage in children from all CP subtypes and levels of the Gross Motor Function Classification System (GMFCS). Methods Analytical cross-sectional study in which 108 children with CP diagnosed by a pediatric neurologist were included with any type of dysfunction and from all levels of the GFMCS. Slaughter equation, Gurka equation and Bioelectrical impedance analysis (BIA) as reference method, were used. Groups were stratified by sex, CP subtypes, GMFCS level and Tanner stage. Median differences, Kruskal–Wallis, Mann–Whitney U test, Spearman's correlation coefficients and simple regressions were used, also multivariate models were performed. Results The Slaughter equation differed from the other methods in the total population and when it was compared by sex, CP subtypes, gross motor function and Tanner stage. The Gurka equation showed significant differences by sex and gross motor function. Gurka equation correlated positively and significantly with BIA to estimate the fat percentage in all the CP subtypes and levels of the GMFCS. Tricipital skinfold (TSF), arm fat area (AFA) and weight for age index (W/A) showed the highest variability with respect to fat percentage. Conclusion Gurka equation is more appropriate and accurate than Slaughter equation to estimate the fat percentage in children with CP from all subtypes and levels of the GMFCS.