In robust nonparametric kernel regression context, we prescribe method to select trimming parameter and bandwidth. Through solving estimating equations, we control outlier effect through combining weighting and trimming. We show asymptotic consistency, establish bias, variance properties and derive asymptotics.