Automatic analysis of the swimming posture of koi is very useful for determining the growth status and ornamental value of these fish. We propose an automatic tracking method based on a particle filter to measure the swimming posture of koi. In water, traditional color-based particle filters are unable to deal with issues such as shadows, reflections and abrupt motion, a novel matching function was investigated for the particle filter for tracking koi motion in water. First a particle filter with a traditional color histogram matching function is used for tracking. Then a new matching function based on a center-surrounding cue is applied to amend the particle distribution and improve the tracking robustness in each iteration. We demonstrate the effectiveness and robustness of our approach through experiments and comparisons using some challenging video sequences of koi swimming.