Among the forgotten lessons that good architects are slowly relearning is the glory of the city street. The lesson is simple: The street itself often is the work of art, and not the buildings upon it.

The paradox of this lesson, which runs so contrary to the grain of modern architecture, is that it takes as much imagination and skill to design a building with the street in mind as it does to create the type of attention-getting monument so dear to the modernist tradition -- and dearer still to speculative office builders.

In addition to that, it takes giant doses of subtlety and tact, qualities often in short supply in the large-scale urban architecture of the postwar era. To do the job right the good older buildings already on the street must of course be saved, but the job does not end with preservation. The older buildings also must be respected.

A design by Warren Cox and George Hartman for the Sumner School site at 17th and M streets NW is the latest evidence that local firms are making progress in their efforts to reinvent the wheel of tactful city architecture on a large scale. It is a slyly argumentative design, literally an essay in brick, stone, steel and glass on the complex process of combining new and old buildings, harmonizing disparate architectural styles, and creating an eminently civilized streetscape.

Whether the $40-million project gets built or not is another question, but things look good. The scheme won an architectural/development competition sponsored by the D.C. School Board, which in the very nick of time thwarted an attempt by the District government to raze the landmark Sumner building for reasons, it said, of public safety. Boston Properties and First City Properties, principal partners in the development, are negotiating with the school board and city government concerning terms of the lease. If all goes well, construction could begin next summer.

Even if it were not to materialize, however, the design itself would merit careful study and, in principle if not in detail, emulation. Its most obvious payoff is that it will save the 109-year-old Sumner School and its M Street neighbor, the 94-year-old Magruder School, a building whose destruction seemed inevitable.

Much of the credit for this happy turn of events is due the development team, which had the foresight to seek an option to buy an adjacent piece of property on M Street and thereby guarantee itself sufficient density and profit without erasing the Magruder, neither architecturally nor historically so important as the Sumner building but, still, a fine 19th-century brick schoolhouse. The Hartman/Cox firm took it from there.

The design is at once simple and complicated. Basically, it locates the bulk of a new office building behind and to the sides of the Sumner and Magruder buildings, placing them, in effect, upon a framed platform -- a starring role for which these modestly robust old structures are admirably suited. There is ample precedent for this solution, of course. Locally the most notable example, John Carl Warnecke's federal office buildings framing Lafayette Square, dates back about 15 years.

All the same there is more than meets the eye in the dryly sophisticated Hartman/Cox scheme. Learning a lesson from the too-massive Warnecke red brick towers, the architects disguise the bulk of the office additions with mirrors -- literally and metaphorically.

Literally, they propose to sheathe the upper floors and service penthouse of a nine-story office building, set back more than 90 feet from the sidewalk, with a mirror-glass curtain wall that from a street-level vantage point will reflect nothing but sky. The lower floors of this conventional curtain-wall structure will be sheathed in glass of a darker hue, the better to form an unobtrusive backdrop for the Magruder building. "It's an attempt to make a big building as invisible as possible," Cox explains.

And then there are the wings of the new building, which in each case "mirror the facades of their next-door neighbors," as Hartman says with but slight exaggeration. This means that a two-bay extension of the Sumner School along 17th Street will conform not only to the height and cornice line of the old structure -- about as far as conventional modern architects have been willing to go in adapting new to old -- but will mimic its window designs and lively brick-and-stone textures as well.

On the eastern edge of the site along M Street, upon the narrow parcel of land so wisely picked up by the developers, there is a large jutting wing that picks up its ornamental details and granite textures from the Jefferson Hotel, an appealingly sedate if borderline humdrum Beaux-Arts building on the corner of 16th and M streets. Thus in a single project, Hartman and Cox manage to pay the respect of imitation to two excellent architects -- Adolph Cluss, who designed Sumner in 1872, and Jean Henri de Sibour, who designed the Jefferson in 1923 -- and to clone two extremely dissimilar architectural styles.

Obviously there is an element of nose-thumbing gamesmanship in the Hartman/Cox scheme -- two modernist-trained bad boys out of school doing what is absolutely anathema to modernist dogma. The reticent neutrality of their mirror-image architecture, followed through with rigorous logic in the absolutely mediocre alley faces of their new buildings, are calculated to drive theoreticians crazy.

But the point of the game is entirely serious. The most important thing about the design is that, in most respects, it works. It promises not only to save a couple of attractive if derelict old school buildings, but genuinely to honor them.

The slate roof, tower pinnacle and Victorian Gothic dormer windows of Cluss' Sumner building, so unceremoniously ripped away, will be completely restored. The Magruder School, if anything, gets even better treatment. To make it a true centerpiece for the ensemble (and to allow construction of underground parking) the building will be dismantled brick by brick and reconstituted in a slightly different spot -- shifted precisely eight feet to the west.

As a result, we are promised a lively series of textures, forms and spaces on the north side of M Street instead of a typical downtown megablock. Indeed, given the strong contrasts between the Hartman/Cox scheme on the north and the contemporary ziggurat-style building designed by Skidmore Owings and Merrill for the National Geographic Society on the south side of the street, the 1600 block of M Street promises to become one of the more visually interesting blocks in the city.

The deficiencies of the Hartman/Cox design are relatively minor. On the inside, the skylit entrance atrium will be an odd little affair, especially if, as planned, it is set off-center to the office building's elevator core. This underlines the fact that all this architectural energy was spent to package a no-nonsense office building: pretty to look at, for sure, but basically boring within, by no means a minor flaw to the people who have to work there.

Outside, those stage-set courtyards flanking the Magruder building are beautifully proportioned and appropriately austere, but they can, and should, be made more hospitable. Because the new National Geographic building across the street will slope back so dramatically, the Magruder courtyards will receive the sun for most of the day. Without cluttering these spaces unnecessarily, the architects should be able to find unceremonious ways to invite people to use them.

Still, this is a brilliant, quietly polemical design. Literal copying of bygone styles is not necessarily the best way to address the question of large-scale, urban architectural context. But by stating their solution so clearly, Hartman and Cox provide a vigorous, thought-provoking reintroduction to the potential civility of the city street.