Few issues in the defense establishment have been studied more intensely than the military's software difficulties -- a task force convened last year began its work by reviewing the 230 recommendations contained in a dozen earlier reports on the subject.

These critiques and the comments of dozens of computer scientists, managers and Pentagon officials interviewed for this series point to three main weaknesses in the Pentagon's handling of software projects:

The Pentagon often botches the initial job of defining what it wants software systems to do, in part because of the enormous complexity of the tasks military planners want computers to perform. Many are giant, one-of-a-kind undertakings, requiring years of work and hundreds of people, that build only minimally upon previous knowledge and that can't be tested under the actual conditions in which they will be used.

Though the needs of the military change constantly in reaction to world events, the Pentagon often tries to decide in advance exactly what a software project is supposed to do. "To think that these guys can sit in the Pentagon and come up with all the numbers, all the threats, that they have all the knowledge so cleanly in their heads that they could come up with a specification that ... would satisfy all the needs in the real world -- that is naive," said Frank Druding, director of a software productivity center at Ford Aerospace, recently bought by Loral Corp.

A regimented, bureaucratic approach to decision-making, ingrained in the military, is poorly suited to the development of software, a process that demands experimentation and constant refinement.

In essence, according to Defense Department officials and contractors, the Pentagon is rooted in a system of rigid, step-by-step schedules and neatly drawn lines of authority that are more suited for building tanks or airplanes than software that is designed but never "constructed" in some physical sense.

As the Pentagon and its contractors attempt to cope with the inevitable changes to software -- many of the modifications requested even as the design and code-writing is underway -- projects become inundated with specifications and accounting paperwork, creating endless opportunities for confusion.

Software Productivity Research Inc., a Cambridge, Mass., consulting firm, estimates that the military requires 200 words of description for each line of software instruction that is written -- and each software project could have millions of lines. Paperwork, the company found, represents nearly half the cost of producing military software.

The Pentagon's software projects -- like its more traditional weapons systems -- often are subject to damaging political pressures from Congress and elsewhere in government. Contractors and military project managers say this often forces them to make overly optimistic predictions about when software will be done and how much it will cost, increasing the risk that rules will be disregarded or side-stepped.

"People get under extraordinary pressure {and} start skipping steps to make their dates," said Watts Humphrey, an official at the Pentagon-funded Software Engineering Institute at Carnegie-Mellon University.

Slowly, the Pentagon is modifying some procurement policies to permit greater flexibility in dealing with unforeseen problems in producing software, while at the same time still meeting the Pentagon's need for regimen and control.

Attempts are being made, for example, to let users "test drive" systems well before they are completed to help close the three-way communications gaps that persist among Pentagon buyers, the ultimate users in the field and the firms writing the software.

On the technical front, $150 million is spent annually by the Pentagon on research into ways to instill more efficiency in military-software development and to encourage contractors to reuse certain basic parts of software programs from earlier projects.

But whether any of this is sufficient remains in doubt. A recent survey found many Pentagon suppliers, as well as senior government officials, skeptical that the nation could meet its military software needs over the next five years.

"We're becoming a mutual frustration society," said Lloyd Mosemann II, a deputy assistant secretary of the Air Force who recently assumed responsibility for software policies. "We're not quite sure how to break the logjam."