Considerable efforts have been invested recently to improve electrocardiographic (ECG) classification accuracy for left ventricular hypertrophy (LVH). This study examines how LVH classification accuracy is influenced by (1) the selection of an echocardiographic standard for LVH, (2) LVH severity level in the test groups, and (3) the adjustment of LVH criteria for obesity and age. Using data obtained from large, community-based populations, this study explores prospects for improving ECG models for LVH classification and examines some of the general characteristics of newer ECG models for estimating left ventricular mass (LVM) on a continuous scale. The results indicate that the apparent ECG classification accuracy for LVH is substantially influenced by echocardiographic standards and criteria for LVH, LVH severity level, and selection criteria for test populations, and these differences explain some of the often substantial differences in test results from clinical versus community-based evaluation studies. The low reproducibility of echocardiographic LVM as the standard is a limiting factor in attempts to improve ECG criteria for LVH and LVM prediction models. Adjustment of ECG amplitudes to anthropometric factors that simultaneously influence LVM may result in confounding effects and may lead to the development of inappropriate models. The performance of ECG models for LVM prediction improved substantially by the inclusion of body weight as a covariate with ECG variables. The addition of standing height and various covariates reflecting obesity did not improve LVM prediction accuracy. Compared to the older LVM prediction models of the Novacode ECG program, the correlation between echocardiographic and ECG estimates of LVM increased sufficiently (from 0.33 to 0.54 in women and from 0.46 to 0.62 in men) to suggest that these improved ECG models are suitable for monitoring LVH progression/ regression in study groups participating in hypertension intervention trials.