It has been shown that fluid intelligence (gf) is fundamental to overcome interference due to information of a previously encoded item along a task-relevant domain. However, the biasing effect of task-irrelevant dimensions is still unclear as well as its relation with gf. The present study aimed at clarifying these issues. Gf was assessed in 60 healthy subjects. In a different session, the same subjects performed two versions (letter-detection and spatial) of a three-back working memory task with a set of physically identical stimuli (letters) presented at different locations on the screen. In the letter-detection task, volunteers were asked to match stimuli on the basis of their identity whereas, in the spatial task, they were required to match items on their locations. Cross-domain bias was manipulated by pseudorandomly inserting a match between the current and the three back items on the irrelevant domain. Our findings showed that a task-irrelevant feature of a salient stimulus can actually bias the ongoing performance. We revealed that, at trials in which the current and the three-back items matched on the irrelevant domain, group accuracy was lower (interference). On the other hand, at trials in which the two items matched on both the relevant and irrelevant domains, the group showed an enhancement of the performance (facilitation). Furthermore, we demonstrated that individual differences in fluid intelligence covaries with the ability to override cross-domain interference in that higher gf subjects showed better performance at interference trials than low gf subjects. Altogether, our findings suggest that stimulus features irrelevant to the task can affect cognitive performance along the relevant domain and that gf plays an important role in protecting relevant memory contents from the hampering effect of such a bias.