INTRODUCTION: The Global Burden of Disease (GBD) disease classification system, based on ICD-10, is mostly based on hospitalised patients and medically-certified causes of death in developed countries. However, most deaths in developing countries occur at home, without medical attention. Classification systems specific to verbal autopsies are thus needed. The GBD includes ICD codes that cannot be applied to community deaths which often lack detailed clinical information and proportionately distributes ill-defined ICD codes among disease groups without explicit guidelines. We propose a modified classification system–Million Death Study mortality classification system (MDS–MC)—that is practicable for use in verbal autopsy studies worldwide. METHODS: We use a Delphi method to categorise ICD codes into disease categories. These categories are then structured into broader disease groups. We compare our classification system against GBD and check for age and sex-wise epidemiological plausibility of major diseases using the MDS study results for 123 000 deaths from 2001 to 2003 across India. RESULTS: 31 disease groups are divided into four broad disease groups—communicable, non-communicable, injury and ill-defined. The MDS–MC produces built-in quality indicators (such as proportion of ill-defined causes for deaths before old age, range checks, etc) that make it easier to manage the inherent misclassification in verbal autopsy. Major disease profiles are epidemiologically plausible for age and sex. A simple to use web-program enables its use in various settings. CONCLUSIONS: The MDS–MC, designed specifically for coding of verbal autopsies, is a widely practicable classification system for use in developing countries.