How Do You Handle Missing Candles in Historical Data?

Missing observations can affect chart patterns and automated analysis. If you notice gaps in your dataset that do not appear to reflect actual market activity, how do you investigate and handle them?