OBJECTIVE: Connectivity networks are crucial to understand the brain resting-state activity using functional magnetic resonance imaging (rs-fMRI). Alterations of these brain networks may highlight important findings concerning the resilience of the brain to different disorders. The focus of this paper is to evaluate the robustness of brain network estimations, discriminate them under anesthesia and compare them to generative models. APPROACH: The extraction of brain functional connectivity (FC) networks is difficult and biased due to the properties of the data: low signal to noise ratio, high dimension low sample size. We propose to use wavelet correlations to assess FC between brain areas under anesthesia using four anesthetics (isoflurane, etomidate, medetomidine, urethane). The networks are then deduced from the functional connectivity matrices by applying statistical thresholds computed using the number of samples at a given scale of wavelet decomposition. Graph measures are extracted and extensive comparisons with generative models of structured networks are conducted. MAIN RESULTS: The sample size and filtering are critical to obtain significant correlations values and thereby detect connections between regions. This is necessary to construct networks different from random ones as shown using rs-fMRI brain networks of dead rats. Brain networks under anesthesia on rats have topological features that are mixing small-world, scale-free and random networks. Betweenness centrality indicates that hubs are present in brain networks obtained from anesthetized rats but locations of these hubs are altered by anesthesia. SIGNIFICANCE: Understanding the effects of anesthesia on brain areas is of particular importance in the context of animal research since animal models are commonly used to explore functions, evaluate lesions or illnesses, and test new drugs. More generally, results indicate that the use of correlations in the context of fMRI signals is robust but must be treated with caution. Solutions are proposed in order to control spurious correlations by setting them to zero.