In this paper, a software confidence evaluation model based on software fault tree analysis and deep auto-encoding network is established to calculate the confidence of rocket fire control software. First, a fault tree is established accroding to the common architecture of rocket fire control software, and a metric set for rocket fire control software is constructed. Second, an autoencoder is used to perform feature dimension reduction and confidence estimation on historical data, so as to calculate top-event occurrence rate, that is, the probability of the software failure. Finally, an example analysis of the propoded method was carried out, and compared the rusults with the results of the traditional exponential model. The case study show that the confidence evaluation model established in this paper is effective, which can be used for the confidence evaluation of rocket fire control software in the engineering development process. And it can also be extended to other types of software.