Traditionally, national statistical institutes prefer design-based estimation methods like the general regression estimator. However, there is more demand for detailed gures, for example on a regional breakdown, but the sample size is too small to compute reliable gures using design based methods. Model based small area estimation (SAE) can be a good alternative. We investigate this for the Dutch investment survey, with investments in tangible xed assets as target variables, for municipalities as areas. As usual for surveys for enterprises, a stratied sample design is applied, the strata are based on the economic activity and size class, based on number of working persons. There are two interesting aspects in this data. The target variable is zero for almost 50% of the sample, as many enterprises do not have investments every year. Furthermore, the distribution of the non-zero values is right-skewed as few enterprises have very large investments. We compare SAE methods which take the properties of the data into account with SAE methods which ignore these properties, and with a design-based method. It is found that SAE can indeed improve the accuracy of the estimates. However, not all model specifications work well. The selection of the optimal specification is a subject for further research.