We commonly think of information as an instrument for better decisions, yet evidence suggests that people often decline free information in non-strategic scenarios. This paper provides a theory for how a dynamically-consistent decision maker can be averse to partial information as a consequence of ambiguity aversion. It introduces a class of recursive preferences on an extended choice domain, which allows the preferences to depend on how information is dynamically revealed and to depart from the standard expected-utility theory. A new notion of ambiguity aversion, called Event Complementarity, exactly characterizes aversion to partial information. Familiar static ambiguity-averse preferences are embedded into the general recursive model, in which conditions for partial information aversion are identified. The findings suggest that Event Complementarity overlaps with yet still differs from the conventional notion of ambiguity aversion.