The temporal clustering algorithm (TCA) has been developed in order to detect irregular, transient functional MRI (fMRI) activation signals when the timings of the stimuli are unknown. Unfortunately, such methods are also sensitive to signal changes caused by motion and physiological noise. We have developed a modified TCA technique, 2dTCA, which can detect multiple different timing patterns within a dataset so that signals of interest can be separated from artifacts and those of no interest. The objective of this work was to further develop the 2dTCA methods and evaluate their performance in simulated functional MRI datasets. Comparisons were made with TCA and a freely-distributed independent component analysis algorithm (ICA). We created two different sets of six computer-generated phantoms with one and two different simulated activation time courses present in 10 regions of interest. The phantoms also contained real subject rigid and nonrigid body motion and noise. Sensitivity of detection, defined as the true-positive activation rate at false-positive activation rates varying between 0.0001 and 0.01, was compared between methods. Additionally, specificity of detection of the irregular, transient signal of interest was assessed by comparing the number of signal time courses detected by each algorithm. The results suggest that the increased sensitivity of 2dTCA over TCA in detecting this particular signal of interest is comparable to the detection with ICA, but with fewer other signals detected. A few examples of the successful application of 2dTCA to the localization of interictal activity in preliminary studies of temporal lobe epilepsy are also described.