We investigate the possibility to achieve high-fidelity universal two-qubit gates by supplementing optimal tuning of individual qubits with dynamical decoupling (DD) of local 1/f noise. We consider simultaneous local pulse sequences applied during the gate operation and compare the efficiencies of periodic, Carr-Purcell and Uhrig DD with hard π-pulses along two directions (π z/y pulses). We present analytical perturbative results (Magnus expansion) in the quasi-static noise approximation combined with numerical simulations for realistic 1/f noise spectra. The gate efficiency is studied as a function of the gate duration, of the number n of pulses, and of the high-frequency roll-off. We find that the gate error is non-monotonic in n, decreasing as n −α in the asymptotic limit, α ≥ 2 depending on the DD sequence. In this limit πz-Urhig is the most efficient scheme for quasi-static 1/f noise, but it is highly sensitive to the soft UV-cutoff. For small number of pulses, πz control yields anti-Zeno behavior, whereas πy pulses minimize the error for a finite n. For the current noise figures in superconducting qubits, two-qubit gate errors ∼ 10 −6 , meeting the requirements for fault-tolerant quantum computation, can be achieved. The Carr-Purcell-Meiboom-Gill sequence is the most efficient procedure, stable for 1/f noise with UV-cutoff up to gigahertz.