Multiqubit arrays can be prepared in any state by adjusting the parameters in the Hamiltonian that governs their time evolution. In this work we show that these arrays can be used for pattern storage and recall, using techniques of machine learning. Any shape of a character or letter can be encoded as a collection of line segments, represented by the pairwise entanglement between neighboring qubits. We did this in two ways: first, by training the real time evolution; and second, by training to a ground state as a quantum annealing (QA) algorithm, by lowering the effective temperature of the system. In the real time training we succeeded in creating the letters X, M, N, and O, in a four-qubit system, with 2.09 % RMS error. With the QA we trained fifteen different characters on a four- five-, and then six-qubit system, with comparable RMS errors. We also showed that the pattern storage was robust to both classical and quantum noise.