Abstract: The key technology to improve the precision of the cost estimate model is how to pre-process and diagnose the raw cost data.The raw cost data were modified with the value of time and learning curve effect,and then diagnosed with singular value decomposition and variance decomposition ratio for multicollinearity.Hat matrix and eliminated t-variance were adopted to distinguish the value out of the ordinary from independent variable and dependent variable respectively.The influential cases were discriminated with Cook’s distance.The data were guaranteed to fit the demand of the model,and the precision of this model was improved.