Combining theory, models and simulation, computational materials science has become the third leg of materials discovery and development, along with materials synthesis and characterization. While the analysis of errors is quite well developed for experiments, such analysis for simulations, particularly for simulations linked across length and time scales, is much less advanced. In this talk we will discuss the two extremes of the types of multiscale simulations, sequential multiscale and concurrent multiscale, and the role that uncertainty quantification (UQ) can play in each. We will base the discussion on specific examples from our work and the rather sparse literature on UQ in materials simulations. We will end by identifying needs for conceptual advances, needs for the development of best practices, and needs for specific implementations.