This paper addresses the problem of load identification in nonintrusive monitoring application when using load signatures in transient signals. The study presents a systematic approach to design a set of wavelet filters to be used as matching patterns for load identification in nonintrusive monitoring. The effect of different higher order wavelet filters and the signal length on the classification accuracy are investigated. The concepts of wavelet clustering and wavelet-signal matching are introduced and are developed in this paper. Machine learning classifiers are used to automate the load classification process using co-testing. The introduced concept is evaluated on a real test system and the results of the experimental work have shown that 1.5 cycle may suffice to achieve a 97.25% classification accuracy. Moreover, the results have shown that the proposed approach is effective in detecting loads with more electronic components and not previously categorized in the assembly.