The invention discloses an index extreme value based rapid matching method of spectrum characteristic variable. The method is characterized in that a simple linear model is treated as a spectrum correction prediction model for selecting valid characteristic variable; an unitary linear regression model is built for each variable (wavelength) in a spectrum variable set based on the law of lambert-beer; unitary characteristic variables corresponding to the peak value and the valley value are selected based on the target of model prediction index extreme value, and a second variable which can realize perfect matching with each unitary characteristic value is further found out to form a discrete characteristic variable set; and a spectrum correction model is built through the characteristic variable set and can be used for effectively overcoming the problem of spectrum collinearity which often occurs in a simple linear model. The method has the advantages of being simple and convenient in model, little in calculation workload, and high in freedom of selection; the method can be popularized and applied to the fields of raman, near infrared, intermediate infrared and ultraviolet spectrum analysis; and effective technological support is provided to the design of a small portable dedicated spectrograph.