Aim: To determine which standard and non-standard measurements exhibit sexual dimorphism, to develop statistical models for sex estimation using single and multiple measurements, and to compare their accuracy. The additional aim was to test the similarity of sexual dimorphism with other populations from a related temporal and spatial frame, i.e., to test the broader applicability of the developed discriminant functions. Methods: The sample comprised skeletons from two archaeological sites from Salona necropoles, Solin-Smiljanovac (974 graves) and Bencunuše-Porta Salonae (15 graves), dated 1st – 6th and 2nd – 5th century. To develop sex estimation standards, we selected only adult skeletons form skeletal burials and single-burial graves or those in which persons could be separated. We also excluded those with systematic pathological changes or anomalies. For sex estimation, we employed DSP method using 4 – 10 pelvic measurements. We took 84 standard and 19 non-standard measurements on the bones free of post-mortem damage, pathological or traumatic changes. For paired bones, we took measurements on each side to test bilateral asymmetry, and for 20 skeletons we took measurements twice to test repeatability. We tested if skeletal measurements differ in the sample from the classical period and late antiquity. We examined which measurements show sexual dimorphism, developed functions for sex estimation using linear discriminant analysis, and calculated data on accuracy. We compared skeletal measurements to those of late antique Tarragona population, and tested applicability of discriminant functions developed on Tarragona population on our sample. Results: DSP method estimated sex for 106 male and 101 female skeletons with similar age distribution. We excluded nine measurements due to measurement error. Total of 10/48 paired measurements showed statistically significant bilateral asymmetry. Skeletal measurements did not differ significantly on samples from the classical period and late antiquity. 62/80 measurements showed statistically significant sexual dimorphism. We developed 75 univariate discriminant functions with accuracy 63% – 92.1% and 18 multivariate functions with accuracy 77.2% – 93.1%. 284 For 43/75 univariate and 17/18 multivariate functions sexing accuracy was greater than 80%. Articular surface measurements and non-standard shaft circumference measurements reached the highest accuracy rates and were also the most represented in the sample. Univariate discriminant functions that included bilaterally asymmetric measurements obtained higher accuracy when using right-side measurements. Measurements from Salona population were greater than those of Tarragona population and previously developed discriminant functions could not be applied to the sample from Salona. Conclusions: Most of the standard and non-standard measurements were sufficiently repeatable and exhibited sexual dimorphism, which means that standard and selected non-standard measurements are suitable for application in osteometric research. As skeletal measurements from the classical period and late antiquity did not differ significantly, discriminant functions should be applicable on skeletal burials from both periods. In this research, we developed a large number of highly accurate discriminant functions which highlighted the importance both of the standard and of the selected non-standard measurements in sex estimation research. Most of the measurements with the highest accuracy rate were also the most represented once in the sample (especially non-standard circumference measurements), so reliable sex estimation on observed population should be almost independent of skeletal preservation. The research emphasized the importance of the independent analysis of measurements that showed statistically significant bilateral asymmetry because we obtained higher accuracy rates when using functions that included measurements from a greater side. We obtained the highest accuracy with multivariate discriminant models, but their accuracy was not considerably greater than the accuracy of best univariate functions. The population comparison showed that sexual dimorphism was not similar in Salona and Tarragona population and that discriminant functions are not mutually applicable. Therefore, the reliability of discriminant functions developed in the present study should be tested prior to the application to another population.