BackgroundElectrocardiogram (ECG) is commonly used clinically due to convenience, but its accuracy is insufficient for left ventricular hypertrophy (LVH) diagnosis. In this study, we attempted to improve diagnostic accuracy of LVH by establishing models with ECG parameters.MethodsEighty hundred and twenty eight patients were recruited in the present study which were divided into groups according to gender, age and body mass index (BMI). The sensitivity, specificity, Youden index, positive predictive value, negative predictive value and accuracy were calculated using ultrasonic cardiogram criteria of LVH as the gold standard. Area under the curve was also calculated to assess the diagnostic accuracy of 22 conventional ECG criteria in different groups. Stepwise discriminant analyses were performed to establish models of ECG for LVH.ResultsThe diagnostic accuracy of ECG11 (S V2 + R V5,6) and ECG12 (S V1,2 + R V5,6) was significantly higher than the other 20 criteria, while ECG15 (R V5/R V6) was lowest. The ECG12 sensitivity for males was 52.5%, for <60 years old was 44.2%, and for BMI <25 kg/m2 was 46.2%,higher than for females (27.5%), for ≧60 years old (35.7%), and for BMI ≧25 kg/m2(27.6%), respectively. The difference between genders was the most obvious. Based on these observations, the following models for males and females were established:andrespectively. The sensitivities of the two new models were 71.4% and 75.8%, significantly higher than the22 conventional ECG criteria.ConclusionTwo models developed based on gender can be considered for use to investigate the preliminary assessment of the probability of LVH.Electronic supplementary materialThe online version of this article (doi:10.1186/s12872-017-0637-8) contains supplementary material, which is available to authorized users.