Gardening: An eternal spring with my beloved Iris - James Fenton, in the grip of collector mania, outlines how he plans to enjoy his favourite flower in its various forms all year round

I know a poet whose speciality is the cultivation of poppies, but I couldn't follow him down that road. I give space to as many as I can, but Meconopsis does not like my garden, and if you miss out on the Himalayan blue poppy you can hardly say that you specialise in the poppy family.

It is a great family: not just the papavers and meconopsises, but the argemones, corydalises, macleayas, romneyas and the Californian poppies. That's a good group for anyone with a collector's cast of mind.

Not every botanical family would yield such pleasurable results. Some are unsuitable (for example Lemnaceae - the duckweeds) for all but the extrovertly batty. Some are too large numerically. I collect roses, but not Rosaceae (nobody could make a collection of any group so big): Rosa, Prunus, or Malus (all members) would each be challenge enough as a genus. And as for the Compositae, forget it: there are more than 900 genera of daisies.

Hideous though the name sounds, on the other hand, a garden devoted to Scrophulariaceae might well have its charm. It would feature a paulownia (a foxglove tree) with . . . foxgloves underneath. There would be hebes for shrubs, and herbaceous veronicas, snapdragons, mulleins, penstemons and many more fine garden plants.

This may sound a cerebral approach to gardening. In fact it is more visceral. A certain plant grabs you, and you want to know about it in all its forms. I once tried to buy a house from an old man who collected bromeliads (the pineapple family). He seemed to want to sell, but the bromeliads, in their various ingenious polythene homes (there was a whole basement full) had him in their grip.

Irises are my third great craze. The first (during my allotment years) was vegetables; the second was roses; and overlapping with the roses came the irises. In the early stages of the layout of my present garden I took care to find as many species as I could. This year, my fifth here, I began to see that there were several which had either outgrown or been overshadowed in their situation, and I realised the truth of the advice that a bearded iris should never be overshadowed: the tubers must bake, for flowers to be produced the following year.

There is a particular pleasure in this fifth-year phase of a garden begun from scratch, when all kinds of things are paying off. Plants bought in ones or threes have now become clumps. Snowdrops lifted from the borders earlier this year have proved such good multipliers as to encourage one to buy small quantities of other varieties, and be prepared to wait.

The irises, though, which were originally planted all over the place, wherever conditions seemed propitious, would when divided begin to exhaust the stock of appropriate places. If the bearded irises were considered suitable for the front of the borders, before long they would fill them.

There remained the thought of creating a new iris bed, or (as the design of the garden suggested) two new iris beds, on the lawn. But then these would become the focal point of the whole garden. The stars of the beds would be the tall bearded varieties. These flower in May and June. What would the bed be like for the rest of the year?

Iris books encourage one to see the year-round possibilities of the genus. You should be able to have an iris 'in bloom or beauty' all the time. But the 'or beauty' part of that formulation is meant to cover the seedpods of I foetidissima (October to February) while the greatest winter bloomer, I unguicularis, is not normally considered suitable for an open bed.

There came, however, the thought that these iris beds might perk up no end if they were treated like the order beds of a botanical garden, and if they were to accommodate not just irises but the Iridaceae. Among many other genera, this would add crocuses to the scheme, and crocuses alone would contribute to the trick of achieving a ver perpetuum - an eternal spring with borders that were never out of bloom.

The year would begin with those small bulbous jobs, the reticulata irises that begin in January, moving smoothly on to spring crocuses and the hardy Juno irises (I aucheri, I bucharica and I magnifica). By April, the absurdly formal procession of the bearded irises begins in order of height: dwarfs first, tall fellows last.

May, June and July are the periods of iris glut. In August the beds would turn orange, red and yellow as the crocosmias kick in, in their many varieties. September would see the first of the true autumn crocuses (as opposed to the colchicums, which are Liliaceae), and these would carry the baton through to November. Nor need December be a cause for utter despair. The great authority, Mr Bowles, had some I unguicularis in open beds, and says that they merely flowered a bit later than they otherwise would.

Stop, stop] I hear you say. This is all wrong. Where's the variety of foliage? Where's the structure, the architecture of the border? What gives height? What flops over what, after it has died back, in order to conceal a gap? What weaves its way through what? What climbs up what, in this scheme?

The answer is that there is no 'variety', but there is variation of foliage and flower. Everything is a variation on a theme. And the total absence of climbers, floppers and weavers is particularly desirable in the general context of this group of plants.

Another part of the answer is that these objections come too late. The order beds, or the order borders (since we are taking a botanical category and converting it for decorative use), have already been cut in the lawn - two of them, running parallel, each a good three paces wide and 25 long. Sharp sand and grit have already been dug in.

As it happens, the lawn is laid with land drains, several of which pass under the order borders, so that drainage should be almost as good as in a raised bed. Whether this makes things too dry for Schizostylis we shall have to see. There are many, many plants on the borders of hardiness, contained within the 100-odd genera, and some such as Watsonia and Libertia which may do better in a more sheltered spot.

The tall bearded irises form the backbone of the beds. They are set out in groups the shape of a half- closed eye; at a slant, so that the design of the two borders is like a herringbone or a chevron. There are great gaps to be filled in. Various iris species have been placed near the edge, and sisyrinchiums brought in. These are great seeders, and will with luck be the chief weeds of the borders.

For autumn crocuses, an order was sent to Broadleigh Gardens, which does many of the bulbs one would require for a comprehensive hardy Iridaceae bed. For bearded irises, I sent off to Kelways. After these orders, though, and after the gathering in of whatever else may be found in the garden, I shall leave it at that for the moment.

The order borders will be mulched in my favourite way, with a good layer of pea-shingle, taking care not to bury those tubers. So the look of the surface of the bed will imitate the look of those photographs by Roger Phillips or Martyn Rix, of plants in their remote habitats, in some Yunanese waste, or by the shores of Lake Van. And if you say to me, 'There seem to be an awful lot of gaps in your order borders', I shall reply dreamily (having never set foot in the place): 'You know, that's just what Erzurum looks like at this time of year, when the yaks have grazed the hermodactylus.'

Kelways Ltd, Barrymore Farm, Langport, Somerset (0458 250 521). Broadleigh Gardens, Barr House, Bishops Hull, Taunton, Somerset (0823 286 231).

Anna Pavord is on holiday.