I will discuss the reliability of large networks of coupled oscillators in response to fluctuating inputs. The networks considered are quite generic. In this talk, I view them as idealized models from neuroscience and borrow some of the associated language. Reliability is the opposite of trial-to-trial variability; a system is reliable if a signal elicits identical responses upon repeated presentations. I will address the problem on two levels: neuronal reliability, which concerns the behavior of individual neurons (or oscillators) embedded in the network, and pooled-response reliability, which measures total outputs from subpopulations. The effects of network structure, cell heterogeneity and noise on reliability will be discussed. Our findings are based largely on dynamical systems ideas (with a slight statistical mechanics flavor) and are supported by simulations. This is joint work with Kevin Lin and Eric Shea-Brown.