Improving the accuracy of the probability of correct detection under interference conditions remains a pressing task, especially in an environment with dynamic and non-stationary interference. In this context, there is growing interest in using adaptive detection algorithms that can change their parameters following the statistical characteristics of the background noise. One of the key aspects of the synthesis of such algorithms is adequate modeling of interference. Autoregressive models allow for effective modeling of interference using the dependence between the previous values of signals, which is important for optimal interference compensation. The effectiveness of building such models largely depends on the accuracy of estimating their parameters, which directly affects the quality of adaptive interference compensation and, accordingly, the detection characteristics of the general algorithm. Therefore, this article investigates the algorithms for estimating the parameters of AR models - in particular, maximum likelihood methods, recursive and classical Yule-Walker, and Levinson–Durbin approaches. Attention is also paid to studying the impact of the selected estimation algorithm on the accuracy of approximation of the statistical characteristics of the noise background, as well as on the subsequent effectiveness of adaptive signal detection. For this purpose, a two-stage computer simulation was implemented: at the first stage, a comparative analysis of the accuracy of estimates of the parameters of the AR model was carried out; at the second stage, the impact of the obtained estimates on the probabilistic characteristics of adaptive detection in interference conditions was assessed.