Adaptive quadrature methods are usually characterized as either local or global. The methods are distinguished by the scope and nature of the information used to determine when algorithm termination should occur and which intervals should be selected for refinement. By treating quadrature as the generation and search of a tree whose nodes represent intervals, a unified nonalgorithmi¢ framework is developed which encompasses both approaches. This framework provides a method for improving the numerical properties of local adaptive quadrature with respect to termination of intervals by relative error criteria, and total algorithm termination due to excessive function evaluations. The framework also suggests heuristics for interval selection in global adaptive quadrature that bring the actual error closer to the requested tolerance. Extensive computational experience with several heuristics for global adaptive quadrature is described