Survey estimates may be susceptible to the influence of sample units having large design weights associated with unusual observed values. Especially in smaller samples, these sample units can influence estimates disproportionately causing them to be very unstable. In this paper, we consider several model-based approaches for weight smoothing where the design weights are modeled as a function of observed survey quantities. Using these modeled weights, one hopes to reduce volatility in the weights, thus producing better estimates. In this paper we extend prior work on the Current Employment Statistics Survey (CES). Several prospective models are used for the weights, including LOESS curves and Bayesian methods. The new "smoothed" weights are then used to create new survey estimates and we compare these estimates to the true value. Analysis of the fitted weights is performed in the end to find cases where "smoothed" weights may give worse estimates.