The literature on idioms often talks about an “idiom advantage,” such that familiar idioms (spill the beans) are generally processed faster than comparable literal phrases (burn the beans). More recently, researchers have explored the processing of idiom modification and while a few studies indicate that familiarity benefits the processing of modified forms, the extent of this facilitation is unknown. In an eye-tracking study, we explored whether familiar idioms and modified versions with 1 or 2 adjectives {spill the [spicy, (red)] beans} are processed faster than matched literal phrases {burn the [spicy, (red)] beans} when both were preceded by a biasing context. The results showed that adjectives inserted in idioms induced longer fixations and were more likely to elicit a regression. However, idiom verbs and final words were processed with the same ease in all adjective conditions, implying that modifying idioms did not impede their processing. In contrast to the widely reported “idiom advantage,” the results demonstrated that canonical and modified idioms were slower to read relative to matched literal controls. This was taken to reflect the competition between an idiom’s literal and figurative meaning, and subsequently the need to select and integrate the contextually appropriate one. In contrast, meaning integration in literal, unambiguous phrases was easier. We argue that processing costs associated with meaning selection may only manifest when idioms are preceded by a biasing context that allows disambiguation to occur in the idiom region, and/or when literal control phrases are contextually appropriate and carefully matched to idioms. Thus, idiom recognition/activation may elicit the well attested idiom advantage, while meaning selection and integration may come at a cost, and idiom modifications may simply add to the cognitive load.