We revisit the problem of designing sublinear algorithms for estimating the average degree of an $n$-vertex graph. The standard access model for graphs allows for the following queries: sampling a uniform random vertex, the degree of a vertex, sampling a uniform random neighbor of a vertex, and ``pair queries'' which determine if a pair of vertices form an edge. In this model, original results [Goldreich-Ron, RSA 2008; Eden-Ron-Seshadhri, SIDMA 2019] on this problem prove that the complexity of getting $(1+\varepsilon)$-multiplicative approximations to the average degree, ignoring $\varepsilon$-dependencies, is $Θ(\sqrt{n})$. When random edges can be sampled, it is known that the average degree can estimated in $\widetilde{O}(n^{1/3})$ queries, even without pair queries [Motwani-Panigrahy-Xu, ICALP 2007; Beretta-Tetek, TALG 2024]. We give a nearly optimal algorithm in the standard access model with random edge samples. Our algorithm makes $\widetilde{O}(n^{1/4})$ queries exploiting the power of pair queries. We also analyze the ``full neighborhood access" model wherein the entire adjacency list of a vertex can be obtained with a single query; this model is relevant in many practical applications. In a weaker version of this model, we give an algorithm that makes $\widetilde{O}(n^{1/5})$ queries. Both these results underscore the power of {\em structural queries}, such as pair queries and full neighborhood access queries, for estimating the average degree. We give nearly matching lower bounds, ignoring $\varepsilon$-dependencies, for all our results. So far, almost all algorithms for estimating average degree assume that the number of vertices, $n$, is known. Inspired by [Beretta-Tetek, TALG 2024], we study this problem when $n$ is unknown and show that structural queries do not help in estimating average degree in this setting.