Quantile regression provides a systematic and robust way of examining the dependence of the response variable on covariates.However,since quantile regression does not assume a parametric error distribution,Bayesian inference of quantile regression requires special treatment.In this paper,we propose a nonparametric hierachical Bayesian quantile regression model based on empirical likelihood.The model uses the ``spike-and-slab" prior to perform variable selection and parameter estimation simultaneously and an efficient Markov Chain Monte Carlo(MCMC)algorithm based on the Laplace approximation to the full conditional distribution is developed.We further prove that the variables selected by the model are asymptotically correct under certain regularity conditions.Simulation studies demonstrate that the proposed model outperforms the linear regression model in most simulation scenarios.