This study aimed to investigate the selection potential for early body mass in a layer resource population, and to compare three inference methods based on their efficiency and accuracy, including restricted maximum likelihood method (REML), Markov chain Monte Carlo method (MCMC), and integrated nested Laplace approximations (INLA). The genetic parameters were estimated by animal models. The layers were collected from a resource population, which were set up by Dongxiang blue-shelled layers crossbred with White Leghorn layers. The total amount of layers, made up of three generations, were 5 405. The phenotypic variance was dissected with the REML and Bayesian approaches. The results showed: 1) The heritability on body mass of the five-week-old layers ranged from 0.49 to 0.59, depending on the estimation method. 2) The MCMC method seemed a slow and low accuracy for genetic evaluation. 3) The use of INLA method could calculate the standard error for variance components, without approximations or use of normality assumptions. 4) Due to the advantage of computation time and accuracy, REML remained the best practical choice for the genetic evaluation in the poultry production. Considering the abundance of genetic potential in the resource population of five-week-old layer, it is better to select on early body mass with animal model.