This paper presents linguistic cloud models for knowledge representation and uncertainty handling in KDD.Multi-dimensional cloud models are introduced as the extension of one-dimensional ones.The digital characteristics of linguistic clouds well integrate the fuzziness and randomness of linguistic terms in a unified way.Conceptual hierarchies based on the models can bridge the gap between quantitative knowledge and qualitative knowledge.In order to discover strong association rules,attribute values are generalized at higher concept levels,allowing overlapping between neighbor attribute values or linguistic terms.And this kind of soft partitioning can mimic human being's thinking,while making the discovered knowledge robust.Combining the cloud model based generalization method with Apriori algorithm for mining association rules from a spatial database shows the benefits in effectiveness,efficiency and flexibility.