Background and purpose: The ideal stroke classification system needs to have validity, high reliability and applicability among different stroke research settings. The Chinese Ischemic Stroke Subclassification (CISS) and the Subtypes of Ischemic Stroke Classification System (SPARKLE) have emerged recently but have not been tested using agreement analysis. As a result, the objective of this study is to investigate the level of agreement among stroke subtype classifications using CISS, SPARKLE and Trial of Org 10172 in Acute Stroke Treatment (TOAST). We also analyse the inter-rater reliability of CISS. Methods: The data include 623 inpatients who have had an ischaemic stroke, accrued from Beijing Tiantan Hospital between 1 October 2015 and 19 April 2016. According to the diagnostic standards of the three subtype classification systems, 299 inpatients who satisfied the requirements of our study were independently classified with etiological subtypes, and we compared the three subclassifications. Results: There was substantial overall agreement among the three classification systems: CISS versus SPARKLE (kappa value=0.684, p<0.001), CISS versus TOAST (kappa value=0.615, p<0.001) and SPARKLE versus TOAST (kappa value=0.675, p<0.001). The inter-rater reliability of CISS was excellent (kappa value=0.857, p<0.001). Furthermore, among the three subtype classification systems, the variance analysis results of the etiological subtypes were not uniform. Conclusion: There were generally substantial agreements among three ischaemic stroke etiological classification systems. CISS is a valid and reliable classification system, with which different stroke research centres can apply and compare data. © Author(s) (or their employer(s)) 2019. Re-use permitted under CC BY-NC. No commercial re-use. See rights and permissions. Published by BMJ.