In order to improve indoor localization accuracy and realize three dimensional positioning, a high precision three dimensional positioning system for visible light communication based on ant colony algorithm is proposed. This system adopts Code Division Multiple Access (CDMA) modulation technology and solves the inter symbol interference between multiple signal sources in visible light communication. In the system, the ID information related to the geographical position of LED signal sourses are loaded directly to the LED drive circuit through direct spread spectrum modulation. After the optical signal is amplified, filtered and sampled, the ID information and the intensity attenuation information can be recovered according to the orthogonality of the spread spectrum code in the Code division multiple access modulation technology. After calculation, the signal intensity attenuation factors are obtained from different LEDs and the global searching property of ant colony algorithm is used to determine the optimum location point. The error correction factor is introduced and the parallel search of ant colony algorithm is used to correct the deviation of the intensity attenuation factors.The simulation results show that the positioning accuracy of the algorithm are 2 cm, 4 cm, 8 cm respectively when the condition of signal-to-noise ratio are 30 dB, 20 dB and 10 dB. When the computation accuracy is higher than 45 cm, the searching efficiency of the ant colony algorithm is higher than that of the ergodic method. After correcting the attenuation factor of light intensity, 100% of test points can reach the positioning accuracy of 5 cm under the signal-to-noise ratio condition of 10 dB. The experimental results show that under the condition of 20 dB signal-to-noise ratio, the positioning error of 92.59% of test points is 8 cm and 96.29% of the test points have three dimensional localization error of less than 10 cm. The maximum localization error is 11.30 cm. After error correction, 96.2% of the test points achieve the precision of 3 cm and 61.6% of the test points achieve the precision of 2 cm. The algorithm reduces the computation of the optimal localization solution and achieves high-precision positioning.