For the randomness and fuzziness problem of condition level boundary information in the power transformers, a new assessment method based on normal cloud theory is proposed. Given a full consideration to the limitation of data use for condition assessment, the limited data collected of the power transformers would be handled to establish the data normal cloud. For the issue of power transformers level boundary ambiguity, it proposes to properly expand each level and establish the level normal cloud. Depending on the different probabilities of cloud droplets in clouds, the cloud association degree is then calculated between the data normal cloud and the level normal cloud. Secondly, the operation data of power transformer is represented the range form. As for volatility for each interval data of performance indicators, it calculates the variance and average interval data and gives different weights index. The result of condition level of the power transformers is finally obtained. Through the analysis of a power transformer station operating data, the simulation example demonstrates that proposed method is efficient and practical to apply to assess the power transformers. This work is supported by National Natural Science Foundation of China (No. 651277023).