The concept of bias is familiar to linguists primarily from the literature on questions. Following the work of Giannakidou and Mari (Truth and Veridicality in Grammar and Thought: Modality, Mood, and Propositional Attitudes, University of Chicago Press, Chicago, 2021), we assume "nonveridical equilibrium" (implying that p and ¬p as equal possibilities) to be the default for epistemic modals, questions and conditionals. The equilibrium of conditionals, as that of questions, can be manipulated to produce bias (i.e., reduced or higher speaker commitment). In this paper, we focus on three kinds of modal elements in German that create bias in conditionals and questions: the adverb "wirklich" 'really', the modal verb "sollte" 'should', and conditional connectives such as "falls" 'if/in case'. We conducted two experiments collecting participants' inference about speaker commitment in different manipulations, Experiment 1 on "sollte"/"wirklich" in "ob"-questions and "wenn"-conditionals, and Experiment 2 on "sollte"/"wirklich" in "wenn"/"falls"/V1-conditionals. Our findings are that both "ob"-questions and "falls"-conditionals express reduced speaker commitment about the modified (antecedent) proposition in comparison to "wenn"-conditionals, which did not differ from V1-conditionals. In addition, "sollte"/"wirklich" in the antecedent of conditionals both create negative bias about the antecedent proposition. Our studies are among the first that deal with bias in conditionals (in comparison to questions) and contribute to furthering our understanding of bias.