回归模型设定是一个难度大的重要问题,它一般涉及回归模型(回归方程)形式、(随机)误差项设定两个方面,还与剩余平方和、样本(变量)数据有关.在误差项服从N(0,σ^2)的情形下,回归方程就是回归模型的数学期望.如果模型设定有误,就产生伪回归问题.本文基于最小二乘原理,研究样本数据、剩余平方和、随机误差项三种特殊情形下的模型设定,提出新的设定思想与方法,从而消除伪回归.“,”Specification of regression model is a very difficult problem, which involves generally to forms of regression model (regression equation) and random error, and relates also to sum of squares of residual and sample data. Regression equation is just mathematical expectation of regression model if random error is as the law of normal distribution N (0, σ^2). False regression results when regression model is wrongly specificated. Three new ideas and methods of specification of model based the least squares principle are respectively proposed with special cases of sample data, sum of squares of residual and random error to eliminate false regression in the paper.