This paper proposes a factor tracking strategy that minimizes the variance of the difference in returns between a target portfolio and return-driving factors. We establish and solve single- and multi-factor tracking models, and analyze the properties of the optimal portfolios. We find that the optimal factor tracking portfolios consist of factor mimicking portfolios and the minimum-variance portfolio determined in the traditional mean-variance model. Moreover, we derive the condition under which the multi-factor tracking portfolio is a linear convex combination of the corresponding single-factor tracking portfolios. Using various datasets from the US market, we show that the factor tracking portfolios outperform the equally-weighted portfolio, measured by both risk and Sharpe ratio. The superior performance of the factor tracking portfolios are attributed to these portfolios’ better upside participation and downside protection.