Abstract: Dispersion fuel is widely used in high-temperature gas-cooled reactor (HTGR), accident tolerant fuel, experimental research reactor, naval nuclear power plant and so on. The chord-length sampling (CLS) method can simplify the geometry modeling of dispersion fuel, which can improve the efficiency. However, traditional CLS can only handle the packing of single particle, and has large error when the packing fraction is high. Aiming to solve these two problems, the improve CLS method was developed in reactor Monte Carlo code RMC, and applied to the fully ceramic micro-encapsulated fuel pin case and HTGR fuel pebble with mixed fuel and poison particles. Results show that the proposed method can handle mixed particles with multiple types, and preserve the accuracy of packing fraction, which provide precise and high efficiency for the critical and burnup calculations.