Different from statistic forecasting, judgmental forecasting, beyond statistical technique, is a subjective way that people use intuition and other cues to forecast what will happen. So it has deviation inevitably. Psychological research found that the deviation in judgmental forecasting has two kinds: inconsistency and bias. Inconsistency is a random deviation from the optimal forecast, whereas bias is a system one. The main performance of bias is as follow: (1) Trend damping. In other words, the forecasts lie below upward trend lines but above downward ones. (2) Elevation effect. That is people often overestimate the trend of no trend series in forecasting. (3) People add random noise to their forecasts. According to the lens-model suggested by Brunswik, there are two type inconsistencies in judgmental forecasting: inconsistency of information acquisition and inconsistency of information processing. The deviation in judgmental forecasting is affected by the process of information processing and features of data series and data presentation format. The process of information processing is quasi-rational. It involved both analytic and intuitive processes. The meaning of intuition is produces an answer, solution, or idea without the use of a conscious, logically defensible, step-by-step process, such as heuristics. There are three main heuristics: availability, representativeness, and anchoring-and-adjustment. Harvey suggested that the type of information on which forecasts were based was the primary factor determining the type of heuristic that people use to make their predictions. The irrational information processing also include overconfident in judgment. People are often too confident to their own beliefs and ability and too optimistic to future. This leads to elevation effect. Hindsight bias is the third type irrational information processing. When recalling their own forecasting, People tend to believe that their forecasting is more accurate than it was. To explore the characteristics of data series that affect judgmental forecasting, researchers conducted a large number of experiments using artificial data. They found that: Different trends of series prediction deviation degree different. High seasonality data may deteriorate judgmental performance. Judgmental forecasting often can't deal well with the series of instability. Random noise has decremental effect on forecast accuracy. Large number of historical data points is not good for judgmental forecasting. Short-term forecast is generally more accurate than long term forecast. Graphical vs. tabular data presentation is a factor that effect accurate, but there is no final conclusion about which one is better. In order to find the effective methods to improve the accuracy of judgmental forecasting, psychologists have studied extensively.Harvey and Stewart suggested some principles respectively that can improve reliability in judgmental forecasting. Other researches found that provision of feedback, decomposition, combining and suggestion adoption could improve the accuracy of judgmental forecasting.