Ideas are connected. New ideas are often seen as creative combinations of previous ideas. I study these connections in the context of motion pictures. A network of 4,445 movies is constructed to indicate which movies are similar. I first examine the properties of the network using descriptive and regression analysis; then I develop a model of network formation for counterfactual analysis. It is found that most movies imitate and evolve around a “core” of the more successful movies. In addition, imitation is both conventional and atypical: a new movie usually follows a stream of similar movies yet simultaneously combines atypical elements from movies outside this stream. This atypicality, if well balanced, has a positive effect on the individual movie’s box office. However, I find that, in the long run, atypical combination may lead to a worse collective box-office performance because of the way it changes the market structure. This paper was accepted by Matthew Shum, marketing.