Slicing is a program analysis technique originally developed for imperative languages. This paper discusses slicing of CLP programs. As data flow in such programs is not explicit, the slicing problem requires a specific formulation and specific techniques.
   The paper gives a precise definition of the slicing problem for CLP and presents a simple slicing method based on variable sharing and groundness analysis. Our approach is based on three declarative notions of slice: (1) for a set of constraints (2) for a derivation tree of a CLP program, and (3) for a CLP program. The first of them is a basic one, expressed in terms of logic and used to define the remaining ones. A derivation tree of a CLP program characterizes a state of the computation of a CLP program in terms of the associated set of constraints. We use the notion of slice of a set of constraints to introduce the notion of slice of a derivation tree. A derivation tree is constructed from instances of program clauses. We define the notion of slice of a program in terms of slices of its derivation trees.
   With these definitions we obtain clear links between all three concepts of a slice. They make it possible to derive slicing techniques for derivation trees and CLP programs by refining some straightforward syntactic conditions which can be easily checked. The declarative formulation of the slicing problem leaves space for possible improvements and gives a clear reference basis for proving the correctness of the proposed slicing methods.