Challenges remain in fluorescence reflectance imaging (FRI) in in vivo experiments, since the target fluorescence signal is often contaminated by the high level of background signal originated from autofluorescence and leakage of excitation light. In this paper, we propose an image subtraction algorithm based on two images acquired using two excitation filters with different spectral regions. One in vivo experiment with a mouse locally injected with fluorescein isothiocyanate (FITC) was conducted to calculate the subtraction coefficient used in our studies and to validate the subtraction result when the exact position of the target fluorescence signal was known. Another in vivo experiment employing a nude mouse implanted with green fluorescent protein (GFP)—expressing colon tumor was conducted to demonstrate the performance of the employed method to extract target fluorescence signal when the exact position of the target fluorescence signal was unknown. The subtraction results show that this image subtraction algorithm can effectively extract the target fluorescence signal and quantitative analysis results demonstrate that the target-to-background ratio (TBR) can be significantly improved by 33.5 times after background signal subtraction. Challenges remain in fluorescence reflectance imaging (FRI) in vivo experiments, since the target fluorescence signal is often contaminated by the high level of background signal originated from autofluorescence and leakage of excitation light. In this paper, we propose an image subtraction algorithm based on One in vivo experiment with a mouse locally injected with fluorescein isothiocyanate (FITC) was conducted to calculate the subtraction coefficient used in our studies and to validate the subtraction result when the exact position of the Another in vivo experiment employing a nude mouse implanted with green fluorescent protein (GFP) -expressing colon tumor was conducted to demonstrate the performance of the employed method to extract target fluorescence signal when the exact position of the target fluorescence signal was unknown. The subtraction results show t hat this image subtraction algorithm can effectively extract the target fluorescence signal and quantitative analysis demonstrated demonstrate that the target-to-background ratio (TBR) can be significantly improved by 33.5 times after background signal subtraction.