In general, a quantum measurement yields an undetermined answer and alters the system to be consistent with the measurement result. This process maps multiple initial states into a single state and thus cannot be reversed. This has important implications in quantum information processing, where errors can be interpreted as measurements. Therefore, it seems that it is impossible to correct errors in a quantum information processor, but protocols exist that are capable of eliminating them if they affect only part of the system. In this work we present the deterministic reversal of a fully projective measurement on a single particle, enabled by a quantum error-correction protocol that distributes the information over three particles.Measurements on a quantum system irreversibly project the system onto a measurement eigenstate regardless of the state of the system. Copying an unknown quantum state is thus impossible because learning about a state without destroying it is prohibited by the nocloning theorem [1]. At first, this seems to be a roadblock for correcting errors in quantum information processors. However, the quantum information can be encoded redundantly in multiple particles and subsequently used by quantum error correction (QEC) techniques [2][3][4][5][6][7]. When one interprets errors as measurements, it becomes clear that such protocols are able to reverse a partial measurement on the system. In experimental realizations of error correction procedures, the effect of the measurement is implicitly reversed but its outcome remains unknown. Previous realizations of measurement reversal with known outcomes have been performed in the context of weak measurements where the measurement and its reversal are probabilistic processes [8][9][10][11]. We will show that it is possible to deterministically reverse measurements on a single particle.We consider a system of three two-level atoms where each can be described as a qubit with the basis states |0 , |1 . An arbitrary pure single-qubit quantum state is given by |ψ = α|0 + β|1 with |α| 2 + |β| 2 = 1 and α, β ∈ C. In the used error-correction protocol, the information of a single (system) qubit is distributed over three qubits by storing the information redundantly in the state α|000 + β|111 . This encoding is able to correct a single bit-flip by performing a majority vote and is known as the repetition code [12].A measurement in the computational basis states |0 , |1 causes a projection onto the σ z axis of the Bloch sphere and can be interpreted as an incoherent phase flip. Thus, any protocol correcting against phase-flips is sufficient to reverse measurements in the computational basis. The repetition code can be modified to protect against such phase-flip errors by a simple basis change from |0 , |1 to |± = 1/ √ 2(|0 ± |1 ). After this basis change each individual qubit is in an equal superposition of |0 and |1 and therefore it is impossible to gain any information about the encoded quantum information by measuring a single qubit along σ z . Because the repet...