The brain constantly adjusts perceived duration based on the recent event history. One such lab phenomenon is subjective time expansion induced in an oddball paradigm (“oddball chronostasis”), where the duration of a distinct item (oddball) appears subjectively longer when embedded in a series of other repeated items (standards). Three hypotheses have been separately proposed but it remains unresolved which or all of them are true: 1) attention prolongs oddball duration, 2) repetition suppression reduces standards duration, and 3) accumulative temporal preparation (anticipation) expedites the perceived item onset so as to lengthen its duration. We thus conducted critical systematic experiments to dissociate the relative contribution of all hypotheses, by orthogonally manipulating sequences types (repeated, ordered, or random) and target serial positions. Participants’ task was to judge whether a target lasts shorter or longer than its reference. The main finding was that a random item sequence still elicited significant chronostasis even though each item was odd. That is, simply being a target draws top-down attention and induces chronostasis. In Experiments 1 (digits) and 2 (orientations), top-down attention explained about half of the effect while saliency/adaptation explained the other half. Additionally, for non-repeated (ordered and random) sequence types, a target with later serial position still elicited stronger chronostasis, favoring a temporal preparation over a repetition suppression account. By contrast, in Experiment 3 (colors), top-down attention was likely the sole factor. Consequently, top-down attention is necessary and sometimes sufficient to explain oddball chronostasis; saliency/adaptation and temporal preparation are contingent factors. These critical boundary conditions revealed in our study serve as quantitative constraints for neural models of duration perception.