Bisection is a search algorithm for numerical CSPs. The main principle is to select one variable at every node of the search tree and to bisect its interval domain. In this paper, we introduce a new adaptive variable selection strategy following an intensification diversification approach. Intensification is implemented by the maximum smear heuristic. Diversification is obtained by a round-robin ordering on the variables. The balance is automatically adapted during the search according to the solving state. Experimental results from a set of standard benchmarks show that this new strategy is more robust. Moreover, it is particularly efficient for solving the well-known Transistor problem, illustrating the benefits of an adaptive search.