Mixing services have emerged as a widely used tool in the blockchain ecosystem to protect user privacy. However, the anonymity provided by these services has also been exploited by malicious actors to evade regulations and conceal illicit fund transfers. Consequently, analyzing the anonymity of mixing services has become a critical research focus in blockchain security. This paper studies Tornado Cash, a popular mixing service on Ethereum, and constructs a dataset of linked deposit-withdrawal address pairs. We further conduct an in-depth analysis of potential feature distributions, offering new insights into anonymity analysis. We begin by collecting transaction records from on-chain data and applying heuristic rules and security reports to identify highly credible linked deposit-withdrawal address pairs. Subsequently, we statistically analyze features such as active timezone distributions, transaction time intervals, and label similarities of counterparties. Our results reveal significant differences between linked and unlinked address pairs. Specifically, linked address pairs exhibit stronger concentration in active timezone distributions and transaction time intervals, whereas unlinked address pairs demonstrate more scattered characteristics. Moreover, the similarity analysis of counterparties’ labels indicates that linked address pairs tend to have more consistent transaction preferences. The proposed dataset provides a benchmark for Ethereum mixing services analysis, which can facilitate future research on mixing services tracing and anonymity.