There are two kinds of devices failures - predicted and unpredicted. The first one includes mechanical damages (wearing, overheating); the second one normally includes electronic components failures. Most of the failures can be predicted by internal disk utilities. All disk parameters considered important by manufacturer are stored and can be retrieved by such utilities.
The constant monitoring of these parameters can detect the device's health with great precision. However the problems related to electronic components can arise spontaneously and are hard to predict. The ratio of predicted and unpredicted failures is about 70% to 30%, therefore device monitoring can determine the device's health with good probability and warn about the necessary measures required to prevent data loss.
Normally unpredicted failures happen in the first 30 days of device usage. After this period of time the real device heath can be predicted with high probability by monitoring S.M.A.R.T. parameters.
PC tech guide's page on S.M.A.R.T. (2003) comments that the technology has gone through three phases:
"In its original incarnation S.M.A.R.T. provided failure prediction by monitoring certain online hard drive activities. A subsequent version improved failure prediction by adding an automatic off-line read scan to monitor additional operations. The latest S.M.A.R.T. III technology not only monitors hard drive activities but adds failure prevention by attempting to detect and repair sector errors. Also, whilst earlier versions of the technology only monitored hard drive activity for data that was retrieved by the operating system, S.M.A.R.T. III tests all data and all sectors of a drive by using off-line data collection to confirm the drive's health during periods of inactivity."