Impulsive noise is a non-Gaussian heavy-tailed random process that is encountered in various communication scenarios. If not catered for, a single impulse will corrupt several symbols in an OFDM block. In this paper we analyze the maximum-likelihood (ML) detection error performance of uncoded OFDM. Results are presented for two different models with emphasis on binary and quadrature phase-shift keying (BPSK/QPSK) constellations. As the number of carriers increases, the error performance actually tends towards the Gaussian noise error curve irrespective of the noise impulsiveness. These results provide benchmarks to validate error performance of various schemes in impulsive noise.