The aim of this paper is to increase the knowledge about industrial land and floor space forecasting. From the rare research on this subject it is known that the knowledge about factors that influence demand and techniques that model demand is not available in abundance. However, different kinds of models are used extensively in practice. This paper focuses the difference between of stated and revealed preferences. In planning practices stated preferences are often used to forecast land and floor space demand on a short term basis (approximately 3 to 5 years). The surveys, on which the datasets that were uses in this paper, were conducted with this aim in particular. Because the surveys were held regularly and the response rates are high, individual firms can be followed through time. So it was possible to put together a panel dataset of firms with stated and revealed preferences. The analysis of this dataset shows that in general stated and revealed preferences for property ownership or rent and location type did match very well. However, preferences for rent are realized more often than preferences for ownership. Whether stated preference matched with revealed preferences for land and floor space was more difficult to determine. At the level of the individual firm, there are huge variations in the stated ñ revealed preference ratios. Some firms only realized only 20% of their stated demand, while others realized double the amount of land of floor space (200%) they initially wanted or more. However, the accumulated stated preference for both land and floor space was only slightly above the accumulated revealed preference. This indicates that stated preferences seem to be useful and reliable to predict demand in the near future. However, this conclusion is valid for the group of firms that had the intension to move and actually did relocate. This leaves two other groups out of consideration: ∑ The unexpected stay group. This group of firms had plans to relocate, but did not had plans to relocate but actually did move. If the survey forecast should be accurate the initial situation, preferences and the choices of these two groups should be almost equal to the expected moves group. This means that: ∑ The initial situation of the expected movers group, the unexpected stay group and the windfall moves group are more or less the same (the groups are not homogeneous). ∑ The stated preferences of the expected moves group and the unexpected stay group are more or less the same. ∑ The stated preferences of the expected moves group and the windfall movers are more or less the same. Unfortunately these conditions are not met. The windfall movers differ substantially from the unexpected stay group, while at the same time these two groups also differ substantially from the expected movers group. Also the group of windfall movers is much larger than the search-moved group. Another, complicating factor is that the preferences seem not stable over time, but show considerable changes. Although, this can be due to a composition effect we should be careful because preferences seem not stationary. Also important for the accuracy of a forecast are: ∑ The time period to which the forecast applies. ∑ The state of the economy (during a recession firms postpone relocations). Although it seems that forecasts based on surveys are inadequate, this conclusion is only partially true because with the expected moves group the aggregated stated and revealed preferences of land and floor space are almost equal. This means that future research has to concentrate on two issues: ∑ We have to shift more to the micro-level and longitudinal approaches to understand search and substitution behaviour. Efforts in modelling the demography of firms show that this may be far more complex than the housing market modelling. Still, analytical models applied in housing studies may suit well. It might well be that existing macro models might be improved with micro-analysis. ∑ We have to look deeper in the decision process of moving, postponing or abandoning relocations.