Most rule induction algorithms including those for association rule mining use high support as one of the main measures of interestingness. In this paper we follow an opposite approach and describe an algorithm, called Optimist, which finds all largest empty intervals in data and then transforms then into the form of multiple-valued rules. It is demonstrated how this algorithm can be applied to mining spatial rules where data involves both geographic and thematic properties. Data preparation (spatial feature generation), data analysis and knowledge postprocessing stages were implemented in the SPIN! spatial data mining system where this algorithm is one of its components.