We present a proof of principle demonstration of a quantum key distribution scheme in higher-order d-dimensional alphabets using spatial degrees of freedom of photons. Our implementation allows for the transmission of 4.56 bits per sifted photon, while providing improved security: an intercept-resend attack on all photons would induce an average error rate of 0.47. Using our system, it should be possible to send more than a byte of information per sifted photon. DOI: 10.1103/PhysRevLett.96.090501 PACS numbers: 03.67.Dd, 42.25.Kb, 42.50.Ar Though quantum key distribution (QKD) has become a commercial reality [1], there is still much interest in fundamental research. One topic of fundamental importance is the design of protocols and implementations which increase the bit transmission rate and/or the security of the QKD scheme. It has been pointed out recently that one can achieve both of these objectives by increasing the dimensionality of the system, that is, encoding a random key string in d-dimensional qudits instead of the usual binary qubits [2,3].It is straightforward to generalize the well-known BB84 protocol [4] to qudits [2,3,5], for which it is possible to send on average log 2 d bits per sifted qudit. Higherdimensional qudits are advantageous not only for an increased bit transmission rate, but also increased security. An eavesdropper employing an intercept-resend strategy would induce a qudit error rate of E d 1 2 dÿ1 d , since half the time she measures in the wrong basis, and consequently sends the wrong state with a probability of d ÿ 1 =d [2,3].Experimentally, there are several methods of encoding d-dimensional qudits in photons, including time-bin [2], orbital angular momentum [6], the polarization state of more than one photon [7], and, more recently, position and linear momentum of entangled photons [8,9].Here we provide an experimental demonstration of quantum key distribution using higher-order d-dimensional alphabets encoded in the transverse spatial profile of single photons. Our scheme is based on the standard BB84 protocol [4], in which Alice chooses which state to send based on the value of a random bit a 1 , while her choice of basis is selected using random bit a 2 . A twobasis BB84 protocol using qudits works the same way [2,3], however, Alice sends states according to the value of a random d-level ''dit''. A simple illustration of our scheme is shown in Fig.