I prefer self-checkouts and use them whenever available. However, the interaction design can be unnecessarily difficult depending on the retailer. Oddly, the most annoying instance locally is the implementation at a high-end grocery store. The cart interaction design expects each item to be placed in the "bagging area" one at a time and to weigh the expected amount.

Any time this requirement is not met or it thinks an "unexpected" item has appeared, the system locks up and waits for a store attendant, who is rarely available. This interaction design is a terrible approach as customers obviously have usage patterns that vary from a single rigid pattern. I've tested it extensively out of sheer morbid curiosity and remain shocked at how bad it is. I've also seen other users frustrated to the point of just giving up and abandoning their purchase.

The most frustrating aspect is virtually all of the fundamental broken-ness of the design is from wrong-headed attempts to discourage theft which are clearly ineffective anyway. It's like the system was designed by a paranoid anal retentive accountant who was completely unaware of the last two decades of interaction design best practice. They clearly never surveyed the range of natural user behavior, or if they did, chose to intentionally ignore it.

Meanwhile, the best implementation of self-checkout locally is Wal-Mart. It has little of the inflexible expectations and relatively few failure modes. The theft deterrence is centered on human psychology more than rigid compliance such as having a live camera view of the user's face visible on-screen. I think it's likely quite a bit more effective while being completely non-disruptive. Frankly, I tend to avoid the high-end store due primarily to their aggressively broken self-checkout system.