We present a study on basis set effects in correlated calculations of core-level states. While it is well recognized that the core-level states require using more extensive basis sets than their valence counterparts, the standard strategy has been to use large contracted basis sets, such as the cc-pVXZ or cc-pCVXZ series. Building upon the ideas of Besley, Gilbert, and Gill [J. Chem. Phys. 130, 124308 (2009)], we show that a much more effective strategy is to use uncontracted bases, such as core or fully uncontracted Pople's basis. The physical grounds behind this approach are explained and illustrated by numerical results. We also discuss other cost-saving strategies, such as virtual space truncation and mixed precision execution.