Graphs algorithms and graph-theoretical problems provide a challenging battlefield for the incremental development of proved models. The B event-based approach implements the incremental and proved development of abstract models which are translated into algorithms; we focus our methodology on the minimum spanning tree problem and on Prim's algorithm. The correctness of the resulting solution is based on properties over trees and we show how the greedy strategy is efficient in this case. We compare properties proven mechanically to the properties found in a classical algorithms textbook.