We study individual ability to memorize and recall information about friendship networks using a combination of experiments and survey-based data. In the experiment subjects are shown a network, in which their location is exogenously assigned, and they are then asked questions about the network after it disappears. We find that subjects exhibit two main cognitive biases: (i) they underestimate the mean degree compared to the actual network and (ii) they underestimate (overestimate) the number of frequent (rare) degrees. We then analyze survey data from two `real' friendship networks from a Silicon Valley firm and from a University Research Center. We find, somewhat remarkably, that individuals in these real networks also exhibit these biases. The experiments yield three further findings: (iii) network cognition is affected by the subject's location, (iv) the accuracy of network cognition varies with the nature of the network, and (v) limitations in network cognition have payoff implications.
