Abstract This paper examines the convergence of per capita carbon dioxide (CO2) emissions for 162 countries across the globe covering all income groups (high-income OECD, high-income non-OECD, middle-income and low-income countries) as categorized by the World Bank. The incidence of stochastic convergence is analysed by making use of recently developed wavelet based unit root tests. The existing studies on this subject make use of conventional unit root tests using time series and/or panel data. A serious drawback of these tests is that they analyse the stochastic behaviour of any series in time domain only whereas the stochastic process may behave differently across different frequencies as well. Thus, wavelet methods are an obvious solution as they consider frequency dimension as well in addition to time dimension when analyzing the stochastic behaviour of any data series and hence these methods encompass the conventional unit root testing methodologies by providing a clear and complete picture of the stochastic process. Our empirical findings, based on latest available time series annual data from 1960 to 2010, lend support in favor of convergence among 38 countries including 18 high income OECD countries, 2 high income non-OECD countries, 13 middle income countries and 5 low income countries while for the rest of 124 countries, the CO2 emission series is found to be non-stationary suggesting the divergence in these countries. These findings are in contrast with the most of the existing studies, which may be due to the use of wavelet based unit root tests, that are of course better alternatives. Some policy implications evolve from the empirical findings.