Since objects above absolute zero agree with the Plank law, the objects’ temperature is reflected by the infrared radiation. With fast response and high resolution, temperature estimation based on mid-wave infrared remote sensing could realize the temperature measurement for small, high-speed and touch-free objects. A new optical system for infrared (IR) image-spectrum integration remote sensing was used to acquire infrared emission spectra from different temperatures of metal. With this basis, we extracted four appropriate spectral features which were the center of gravity position, peak position, the value of wavelength λ1 and the value of wavelength λ2 from the training samples. The relationship between temperature and these features was studied. A multiple linear regression model was established to estimate the temperatures from the spectra. The experimental results showed that, the method could distinguish hot objects with obvious temperature differences. The absolute error was less than 30 ℃ in the experimental temperature range. The accuracy was 98% in the range that the measurement error was less than 20 ℃, which was better than the 2% precision of the general system with the complex strict emissivity, atmospheric transmittance, environmental equivalent radiation temperature and some other parameters. This method could measure the temperature of the remote objects in a simple and effective way, and so could expand the application field of temperature estimation based on infrared remote sensing.