"Many administrators, private and public, have decided that the future of plant breeding lies in genomics, relying on claims that molecular genetics has revolutionized the time frame for product development. ‘Seldom has it been pointed out that it is going to take as long to breed a molecular engineering gene into a successful cultivar as it takes for a natural gene’" - Goodman 2002
Traditional breeding essentially consists of the repeated selection of the best individuals of a plant population over time. This can be accomplished by farmers, hobbyists or professionals and ranges radically in sophistication from the inadvertent selection of genotypes that grow best in a given cultivated environment to massive multi-year statistical studies on large pedigreed families grown in multi-location trials. Regardless of the methods used, breeders are unified in their selection of traits based on phenotype and NOT on genotype (with some limited MAS). Breeders generally don't know (or care) why a plant has a certain trait. They just want it to work. This approach is the strategic opposite of genetic engineering, which aims to first understand the specific mechanism of a given trait first so that it can be consciously and directly modified.
New Sources of Germplasm: Lines, Transgenes, and Breeders*
The end of breeding has been repeatedly and falsely prophesied for going on two decades now - yet even after hundreds and hundreds of millions of dollars of research, only a handful of transgenic traits have been successful. To some extent, it's an unfair comparison as applied genetic engineering has been virtually crippled by regulation. Yet all the same, the above paper by Goodman (2002) does a nice job of outlining all the complications that genetic engineering enthusiasts tend to gloss over in their zeal to take crop improvement beyond breeding.
Far and away, I think the most important reason that genetic engineering has not replaced breeding is that overenthusiastic proponents fail to understand how much of the genetics upon which breeding is built remains unknown. Genotypes can perform radically different in very similar environments and genes can perform radically different within similar genotypes (and of the tens of thousands of genes in any crop, we have a hint of only what a few of them do). Breeding succeeds in the face of these tremendous unknowns because it selects blindly based on phenotypic results.
This distinction between consciously engineering a system and improving it by trial and error (generating many, possibly random iterations and just seeing which works best) is a fundamental distinction seen in many fields from drug discovery to mechanical engineering (to evolving computational cars!). Whether you're trying to improve biological, technological or social systems, iterative selection is best when you don't really understand how your system works - but once you do understand it well enough to make accurate predictions, rational engineering allows massive leaps in what you can accomplish.
Currently, our knowledge of plant biology is nowhere near complete enough to allow wholesale engineering - but it does allow us to make some very small, targeted changes that can occasionally have very big effects (e.g. herbicide and pest resistance). Ten years after Goodman's paper, there is still much that we can hope that genetic engineering will accomplish - but breeding will continue to be the bread and butter of crop adaptation and yield improvement for the foreseeable future.** In 2002, he warned:
"Plagiarizing N.W. Simmonds (1991), we can add MAS, genomics, and possibly even transgenics to the bandwagons we have known. These include (but are certainly not restricted to): induced polyploids, haploids, mutations, overdominance, genetic variances harvest index, high-lysine, small tassels, nitrogen fixation, nitrate reductase, somaclonal variation, bracytic dwarfs, leafy hybrids, precision agriculture, high-oil topcrosses, ag chemical/seed synergy.
Simmonds’ observations merit repeating even after more than a decade, “The bandwagon, as it applies to plant breeding, is expensive and damaging. Resources are being diverted from doing genuinely useful jobs to the pursuit of trendy irrelevance; biotechnology is, I think, accelerating the collapse of proper agricultural research."
Clearly this is an overstatement (now that we're in the future). In particular precision agriculture has gone mainstream and genomics and MAS have been paying dividends. All the same, his ending plea to remember that breeding is the core of crop improvement holds true. Private companies haven't forgotten this (for the few crops and market classes they work on), but our public breeding programs (that cover every other crop) are being rapidly hollowed out and dissolved.
I once heard a nice analogy of genetic engineering vs. breeding.*** Emphasis on genetic engineered "traits" are like drop-in widgets for cars: electric starters, GPS, halogen headlights, etc. But breeding is what shapes the chassis, drivetrain and body. It's great to have all those bells and whistles, but they're more useful on some cars than others...
* Goodman, M.M. (2002). New Sources of Germplasm: Lines, Transgenes, and Breeders Memoria congresso nacional de fitogenetica
** Of course my department, Applied Systems Biology, is one of many groups trying to change this...
*** Rabobank presentation from Genomics in Business, 2011