Results from 4 experiments support the view that, regardless of contextual information, when an end-product interpretation of an utterance does not rely on the salient (lexicalized and prominent) meanings of its components, it will not be faster than nor as fast to derive as when it does. To test this view, we looked into interpretations of salience-based (here literal) interpretations and expectation-based (here ironic) interpretations in contexts inducing an expectation for irony. In Experiment 1, expectancy was manipulated by introducing an ironic speaker in vivo who also uttered the target utterance. Findings show that ironic targets were slower to read than literal counterparts. Experiment 2 shows that ironies took longer to read than literals and that response times to ironically related probes were longer than to literally related probes, regardless of context. Experiments 3 and 4 show that, even when participants were given extra processing time and were exclusively presented ironically biasing contexts, the expectancy for irony acquired throughout such exposure did not facilitate expectancy-based compared to salience-based interpretations.