This article illustrates and differentiates the unique role cognitive interviews and think-aloud interviews play in developing and validating assessments. Specifically, we describe the use of (a) cognitive interviews to gather empirical evidence to support claims about the intended construct being measured and (b) think-aloud interviews to gather evidence about the problem-solving processes students use while completing tasks assessing the intended construct. We illustrate their use in the context of a classroom assessment of an early mathematics construct -- numeric relational reasoning -- for kindergarten through Grade 2 students. This assessment is intended to provide teachers with data to guide their instructional decisions. We conducted 64 cognitive interviews with 32 students to collect evidence about students' understanding of the construct. We conducted 106 think-aloud interviews with 14 students to understand how the prototypical items elicited the intended construct. The task-based interview results iteratively informed assessment development and contributed important sources of validity evidence.