Recently, Transformer-based and MLP-based models have emerged rapidly andwon dominance in time series analysis. In contrast, convolution is losing steamin time series tasks nowadays for inferior performance. This paper studies theopen question of how to better use convolution in time series analysis and makesefforts to bring convolution back to the arena of time series analysis. To this end,we modernize the traditional TCN and conduct time series related modificationsto make it more suitable for time series tasks. As the outcome, we proposeModernTCN and successfully solve this open question through a seldom-exploredway in time series community. As a pure convolution structure, ModernTCN stillachieves the consistent state-of-the-art performance on five mainstream time seriesanalysis tasks while maintaining the efficiency advantage of convolution-basedmodels, therefore providing a better balance of efficiency and performance thanstate-of-the-art Transformer-based and MLP-based models. Our study furtherreveals that, compared with previous convolution-based models, our ModernTCNhas much larger effective receptive fields (ERFs), therefore can better unleash thepotential of convolution in time series analysis. Code is available at this repository:https://github.com/luodhhh/ModernTCN.