Field-based detection of 3:2 pulldown in a sequence of digital video fields using a programmable graphics processor is described. The detection is performed using a threshold value to determine equivalence between a pair of fields of digital video data. Furthermore, additional threshold values may be used to control switching into a mode where duplicated fields of digital video data are identified and not displayed and out of the mode where duplicated fields of digital video data are identified and not displayed. Look ahead can be used to detect when to switch into or out of the mode where duplicated fields of digital video data are identified and not displayed, reducing the occurrence of visual artifacts.