The reported studies investigate online processing of taboo words (e.g. shit) and their censored equivalents (e.g. s**t), relative to semantically matched non-taboo words (e.g. junk). Participants' eyes were tracked as they read sentences which contained one of the critical words. In Experiment 1, participants also encountered censored-neutral words, known as masked (e.g. j**k), but in Experiment 2, participants only encountered the taboo, censored, and neutral conditions, thus manipulating the perceptual certainty of censored words. Taboo and neutral words required similar processing time across all reading measures; liberal post-hoc analyses replicated the null effect. With regards to the censored words, Experiment 1 revealed that early word-recognition requirements were similar between censored, taboo, and neutral words, with censored words requiring additional processing time in later sentence integration measures. However, the results from Experiment 2 revealed no differences in reading time between conditions, suggesting that the masked words in Experiment 1 motivated participants to doublecheck the censored words due to their orthographic similarity. After reading all of the sentences in Experiment 2, participants' memory of the sentences was tested. Participants were able to differentiate between whether they encountered a neutral or profane word (i.e. either taboo or censor), but participants were unable to identify the specific profane word that they encountered in the reading task. We argue that the results relating to the taboo words further clarifies language's role within the functional architecture of cognition while the results relating to censorship informs how statistical regularities of language are used to process lexical-semantic information.