Abstract: Using laboratory experiments, we establish a number of stylized facts about the process of inflation expectation formation. Within a New Keynesian sticky price framework, we ask subjects to provide forecasts of inflation and their corresponding confidence bounds. We study individual responses and find that about 40% of the subjects predominately use a rational rule when forecasting inflation and about 35% of agents simply extrapolate trend. Around 5% of subjects behave in an adaptive manner, while the remaining 20% behaves in accordance to adaptive learning and sticky information type models. However, we provide evidence that many subjects do not rely on a single model of expectation formation, but are rather switching between different models. We also study the implications of the design of monetary policy to the expectation formation mechanism. The degree of backward looking formation increases in an environment characterized by excessive inflation variability and expectational cycles. Some policy prescriptions how to design a monetary policy rule that is robust to different expectation formation mechanisms are also discussed.