Experiments involving phase coherent dynamics of networks of spins, such as echo experiments, will only work if decoherence can be suppressed. We show here, by analyzing the particular example of a crystalline network of Fe8 molecules, that most decoherence typically comes from pairwise interactions (particularly dipolar interactions) between the spins, which cause 'correlated errors'. However at very low T these are strongly suppressed. These results have important implications for the design of quantum information processing systems using electronic spins. [4], are a leading candidate for this. In some of these systems (notably magnetic molecules), the individual qubit properties are controlled by chemistry instead of by nano-engineering (the 'bottom-up' approach [5]), with spin Hamiltonians and inter-molecular spin couplings known and controlled to at least 3 significant figures. Spin also possesses other advantages -information can be encoded in the topological spin phase, with no need to move electrons around. Using spins for quantum information will ultimately require (i) detecting and manipulating single-spins, and (ii) understanding and controlling decoherence.Single spins have been detected in a few ingenious experiments [6], but we don't yet have a general-purpose, single-spin detection/manipulation tool, analogous to single atom STM/AFM. Consider, however, an array of spins, each having a low-energy doublet of states whose splitting is easily controlled by a magnetic field. Even without addressing individual spins, one can still demonstrate coherent qubit operation, using external AC fields to promote resonant transitions between levels, and pulse sequences (e.g. spin echo) to manipulate the phase and measure decoherence rates. This approach is well known for room-temperature bulk NMR quantum computing [7]. Here we treat the case of electronic spins which, unlike nuclei, can be highly polarized at low T . We introduce a formalism allowing the description of any set of spin qubits obtained by truncation to low energy of a larger system, showing how the low-T decoherence rate can be dramatically reduced, even for a network of mutually coupled qubits. To be specific, we treat the case where the qubit is obtained by taking an anisotropic highspin nanomagnet [8] with easy axisẑ, subject to a large transverse field H ⊥ (Fig. 1), giving a low-energy doublet of states with easily controllable energy separation, Fig. 2(a)]. To make quantitative and testable predictions, we then calculate the spin-echo decay rate in a network of Fe 8 molecules. This is a clean, crystalline and stoichiometric chemical compound [8,9], where the inter-qubit and the qubit-environment interactions are known accurately, and it should be a good 'benchmark' for quantitative test of the theory. The resonance experimental set-up and the spin states on the Bloch sphere; (c) At T ≪ ∆o/kB only the lowestenergy eigenstate is populated, |S = 2 −1/2 (|Z+ + |Z− ). A short µwave pulse prepares the system in the |Z+ state (π/2 rotation, correspo...