This thesis analyses the way in which ambiguity is cognitively processed, in translation in general and post-editing in particular, drawing inferences from psycholinguistics, bilingualism, and entropy-based models of translation cognition. Conceptually, it assumes non-selective activation of both languages (source and target) in the translation process, and explores how entropy and entropy reduction can theoretically describe assumed mental states during disambiguation. Empirically, it uses a product-based metric of word translation entropy (HTra), and eye-movement and keystroke data from the CRITT Translation Process Research Database, to shed light on how the conceptual understanding of lexical and structural ambiguity may be manifested by observable behaviour. At the lexical level, examination of behavioural data pertaining to a high-HTra item from 217 participants translating/post-editing from English into multiple languages shows that the item tends to result in pauses in production and regression of eye movements, and that the translators’/post-editors’ corresponding scrutinization of the source text (ST) tends to involve a visual search for lower-HTra words in the co-text and, accordingly, a decrease in the average entropy of the activity unit. Regarding syntax, a Chinese relative clause in the machine translation output, which can involve a garden-path effect, is examined in terms of eye movements from 18 participants. Results show that, contrary to monolingual reading, disruptions of processing tend to occur not in the later part of the sentence where the wrong parse is disconfirmed, but in the earlier regions where the most quickly-built analysis is semantically inconsistent with the ST. Structural disambiguation and re-analysis seem to be bypassed. This suggests that, on the one hand, reading for post-editing receives a strong biasing effect from the ST, and on the other, argument integration is more appropriately explained from an incremental processing perspective rather than a head-driven approach, as thematic roles seem to be assigned immediately in reading for post-editing. While the lexical analysis supports a parallel disambiguation model, the structural analysis seems to support a serial one. In terms of translation models, both emphasize the impact of cross-linguistic priming and the presence of considerable horizontality in the translation process.