In this paper, a UAV flight formation technique is proposed for homing to the PM2.5 pollution source described by the Gaussian steady-state dispersion model. Single or two UAVs with PM2.5 detection sensors are difficult to find the exact particle distribution gradient. The gradient is used as information for developing guidance and control laws for detecting the emission source. Different formation manner of UAV teams for the PM2.5 pollution source detection is analyzed. The minimal number of UAVs team and formation manner is proposed and verified by vast flight simulations. Two-dimensional guidance laws are developed for horizontal and vertical homing. The Gaussian steady-state dispersion model can provide an analytical description of the air pollution source for analyses and designs. The proposed method can be applied to another dispersion model easily.