We analyze the application of partial-response equalization and maximum-likelihood sequence estimation in magneto-optical readout. Two filters are proposed, and several aspects of their performance are examined. Filter I has 8 states in its state-transition diagram and is therefore easier to implement than filter II, which has 32 states. We discuss the required signal-to-noise ratio as function of the recorded bit density for these filters. The effects of jitter and bloom on the eye patterns of the output signals are also examined by computer simulation. This analysis indicates that filter II is somewhat superior to filter I, presumably because the output of filter II is more similar to the actual readout signal. We determine the distribution of Euclidean distance between pairs of output sequences and compute upper bounds on the probability of sequence error for both filters. Using two different methods of precoding (i.e., mapping of the user data to the magnetic pattern on the disk), we also compute the probability of bit error for the user data and show that one precoding scheme is slightly better than the other.