Abstract: Nanofluids can be a new type of heat transfer medium with the high thermal conductivity. Particularly, the traditional heat transfer media cannot fully meet the growing needs of thermal management in the electronic devices in various fields. Among them, the stability is a key factor for the wide application of nanofluids. In order to obtain more stable nanofluids, the dispersants were first screened, and then the stability of the Carbon NanoTube (CNT) nanofluids with the selected dispersant was optimized using the Response Surface Method (RSM). CNT nanofluids were prepared by the two-step method. The comprehensive evaluation index was used to compare the effects of four surfactants on the stability and thermal conductivity of CNT nanofluids. The stability evaluation included the particle size potential and absorbance method. Taking the particle size as the evaluation index, three influencing factors were selected as the mass fraction, ultrasonic oscillation time, and placement of CNT nanofluids. A three-factor and three-level Box-Behnken test was designed to optimize the parameters. The RSM was used to investigate the effects of various factors on the interaction of nanofluid stability. The results show that compared with the rest dispersants, the CNTs nanofluids with the DB dispersant presented the smallest particle size, the highest absorbance, the highest thermal conductivity, and the best comprehensive performance, when the ratio of carbon tube to dispersant was 1:2. Therefore, the DB dispersant was selected to evaluate the performance of CNTs nanofluids. A RSM model was established to accurately describe the particle size of CNT nanofluids using the Box-Behnken experimental design. At the same time, the quadratic polynomial regression equation was obtained for the response value particle size (Y) with the mass fraction of CNT nanofluids (X1), ultrasonic oscillation time (X2), and storage time (X3). The quadratic polynomial mathematical model was established by the experimental data. There was the highly significant (P<0.0001) with the correlation coefficient R2=0.964 4, indicating an excellent fit between the predicted and the actual value. It infers that there was an excellent consistency between the model and the experimental data. An optimal combination of factors was achieved for the stability of CNT nanofluids using the quadratic regression model, where the mass fraction of CNT nanofluids was 0.27%, the ultrasonic time was 83.45 min, and the standing time was 8 h. In this case, the average particle size was 121.58 nm, and there was a 0.016% error with the predicted value of 120.60 nm. Therefore, this improved model was feasible to optimize the stability of CNT nanofluids. The actual value was close to the predicted value. There was a certain reference value for a theoretical and experimental basis, in order to improve the stability of nanofluids. Response surface test also revealed that the ultrasonic oscillation time and placement time can be selected to improve the stability in the subsequent preparation of nanofluids.