Graph data is ubiquitous in many domains such as social network, bioinformatics, biochemical and image analysis. Finding subgraph isomorphism is a fundamental task in most graph databases and applications. Despite its NP-completeness, many algorithms have been proposed to tackle this problem in practical scenarios. Recently proposed algorithms consistently claimed themselves faster than previous ones, while the fairness of their evaluation is questionable due to query-set selections and algorithm implementations. Although there are some existing works comparing the performance of state-of-the-art subgraph isomorphism algorithms under the same query-sets and implementation settings, we observed there are still some important issues left unclear. For example, it remains unclear how those algorithms behave when dealing with unlabelled graphs. It is debatable that the number of embeddings of a larger query is smaller than that of a smaller query, which further challenges the remark that the time cost should decrease for a good algorithm when increasing the size of the queries. In this paper, we conducted a comprehensive evaluation of three of most recent subgraph algorithms. Through the analysis of the experiment results, we clarify those issues.