The ability to use numerical evidence to revise beliefs about the physical world is an essential component of scientific reasoning that begins to develop in middle childhood. In 2 studies, we explored how data variability and consistency with participants' initial beliefs about causal factors associated with pendulums affected their ability to revise those beliefs. Children (9-11 years old) and college-aged adults ran experiments in which they generated, recorded, and interpreted data so as to identify factors that might affect the period of a pendulum. In Study 1, several children and most adults used observed evidence to revise their initial understanding, but participants were more likely to change incorrect noncausal beliefs to causal beliefs than the reverse. In Study 2, we oriented participants toward either an "engineering" goal (to get an effect) or a "science" goal (to discover the causal structure of the domain) and presented them with variable data about potentially causal factors. Science goals produced more belief revision than engineering goals. Numerical data, when presented in context, with appropriate structure, can help children and adults reexamine their beliefs and initiate and support the process of conceptual change and robust scientific thinking.