"How how can a new spell be created?"
Officially, as this fine answer mentions, Researching Original Spells covers the basics of creating a new spell: the creature spends both 1 week and 1,000 gp per spell level of the new spell studying in an extensive library, and at the end the researcher makes a Spellcraft skill check (DC 10 + the new spell's spell level). Success means that either the researcher learns the spell or the DM reveals that the proposed spell isn't viable, and failure means that the researcher can start over. (This is on Dungeon Master's Guide (2000) 42 and DMG (2003) 198.)
However, Tome and Blood (July 2001) goes into much greater detail on Researching New Spells (81–3), starting with this mandate:
Spell research is a rare case in which you, your DM, and your character all play a part. Your character must spend time and money working on the spell. You, however, must write the spell description and submit it to your DM for approval. (81)
This is the only place in the game that I'm aware of where the player is told that he must compose something. This makes researching original spells unique in that a player's real-life ability—his writing skill—can have a direct impact on the player's PC's ability to change the game. (Whether this is for good or for ill depends on the table.)
Anyway, from there, Researching New Spells continues, explaining the ins and outs of researching: a researcher can pretty much only research during the weeks she's researching. Stopping research ruins the research, and doing anything else (beyond simple survival) ruins the research. And are you tired of borrowing the extensive library? Assemble your own for 10,000 gp! You'll probably want to put it in something like a portable hole (DMG 264) (20,000 gp; 0 lbs.): a library weighs at least 600 lbs. (see TB 71, 72).
Researching New Spells then discusses the viability of new spells, providing benchmark spells for each spell level. Because Tome and Blood is a Third Edition text and this section was never subject to the 3.5 revision, the DM must make minor adjustments to some benchmark spells (e.g. one 1st-level benchmark miscellaneous spell is spider climb), but, overall, the benchmark spells provide some insight as to one designer's intent of what should be the comparative power level of spells at each level.
Researching New Spells concludes with a Behind the Curtain discussion of how it was decided that two new spells have the levels they have—the 2nd-level Clr, Drd, and Sor/Wiz spell gaze screen [abjur] (TB 90) and the 5th-level Sor/Wiz spell ghostform [trans] (TB 90). (As an aside, when the spell ghostform is updated for the 3.5 revision by the Spell Compendium (Dec. 2005) (103), it becomes an 8th-level spell. Make of that what you will.)
So far as I can tell, researching original spells isn't ever addressed again in detail in the remainder of the game's official corpus. For example, Magic of Faerûn, Complete Arcane, Unearthed Arcana, Dragon Magic, Spell Compendium, Dungeon Master's Guide II, Player's Handbook II, Complete Mage, and Magic of Eberron all lack any discussion of designing original nonepic spells.
"But is Tome and Blood a good source for guidelines on creating new spells?"
It's pretty much as good as it can get. Could there be firmer guidelines? Sure. However, in the end, the guidelines that are provided jibe with the standard 3.5 paradigm of Compare your new stuff to our official stuff, and change your new stuff until you think it's balanced against our official stuff. That makes the information on benchmark spells particularly useful: it's far better than the nothing that the Dungeon Master's Guide provides and illustrates—to some degree, at least—designer intent, which is kind of necessary when discussing adding balanced original material to an existing game.
The unasked question: "Are these good rules?"
The question doesn't ask this, but I feel I must address it anyway. Frankly, I think these rules could be a lot friendlier. As both the price of Original Spell Research and the skill necessary to perform Original Spell Research are trivial, time and access to a library become its nontrivial mechanical aspects. With the scarcity of appropriate libraries (randomly generated towns only have a 5% of being big enough for one, for instance) and Tome and Blood's clarification that Original Spell Research is an all-consuming task, the rules tacitly seem to encourage the passive-aggressive DM to troll players: "Sure, because your PC has the money and the skill, she can totally research original spells," says the DM, "if she can find a library, and if she isn't interrupted for a couple of weeks." Then the DM—who doesn't want to say he just doesn't want original spells in the campaign—runs adventures in small cities and littler towns or has assassins attack the researcher on the last day of his research.
Even more galling, though, is that the DM is supposed to reveal if a spell's not viable only after the original spell research is complete. The player's supposed to write the spell and submit it to the DM, and then the player's PC is to find a library, spend the money, spend the time, and succeed on the skill check, and then and only then is the player to be told by the DM that the spell the player wrote isn't viable. That kind of scenario in real life makes me terribly angry; that the rules encourage duplicating that scenario for fun makes me sick to my stomach. To be fair, the DMG does say that the DM should "feel free to work with the player before the research begins and give him guidance on the parameters under which an original spell might be acceptable in [that DM's] game" (198), and that's a good start, but that the DMG doesn't require that conversation still makes me sad.
Third-party material
I looked at the rules for researching original spells in Encyclopedia Arcane: Chaos Magic (2001) that's suggested by this answer, Quintessential Wizard II (2004) that's suggested by this answer, and Ultimate Magic (2011) for the 3.5 spinoff Pathfinder that's suggested by this answer.
Encyclopedia Arcane: Chaos Magic doesn't present a system for designing original spells per se but a new standard class—the chaos mage—that employs a new magic system that creates magical effects on the fly using a six-step process. The last step has the caster making a chaos-mage-class-level-plus-Charisma-modifier check against a DC informed by the previous five steps. There's no need for an examination of chaos magic in general here; instead, what's important is Crafting Chaos Magic (18–30) that describes how to determine those caster level check DCs. While this reader can imagine a DM adopting Crafting Chaos Magic as the hub of a campaign's method for designing original spells, the DM must still do a fair amount of work to translate its numbers into spell levels and confirm that the original spell's results are balanced.
Quintessential Wizard II on Spell Research (81–92) uses what it calls Level Points to establish an original spell's spell level, with, for example, grander and more impressive effects, more difficult saving throws, and combinations of effects leading to a higher Level Point cost and, therefore, a higher level spell when the research is completed. To be fair, while I'm certain that a skilled accountant could develop devastating spells on the cheap absent oversight with this system—almost any point system is subject to abuse absent oversight—, I didn't examine this in great detail. What is interesting is, since the DMG's Original Spell Research rules are excluded from the SRD for 3.5, Wizard II makes up its own: an original spell has, essentially, a price in raw materials of its spell level squared × 100 gp (minimum 50 gp), and the researcher makes progress toward researching the spell using a variant of the rules for the skill Craft (DC 15 + spell level). To this reader that's an improvement: it's not perfect—the rules for the Craft skill are weird—, but at least it marshals an existing mechanic unlike the official rules.
Ultimate Magic on Designing Spells (128–39) packs into its pages guidelines for creating original spells by comparing original spells to existing spells. While this may seem both really bloody obvious and a rehash of Tome and Blood—and, to some degree, it is, even including another round of benchmark spells—the level of detail Ultimate Magic goes into is much more expansive, far outstripping Tome and Blood. Some Pathfinder-specific elements creep in (e.g. the descriptor emotion), but (ahem) ultimately Magic encourages comparing original spells with spells from Pathfinder's initial rulebook, the spine of which is the SRD for 3.5, at one point Ultimate Magic going so far as to say, "The [Pathfinder] Core Rulebook spells [that are drawn almost entirely from the SRD for 3.5] are the most playtested, optimal versions of spells in the game" (136). Were a player in in one of my campaigns to need guidance beyond Tome and Blood in determining a spell level for an original spell, this is the text I'd point that player to.
There are undoubtedly other sources that provide rules for designing original spells: the absence of the DMG's Researching Original Spells rules in the SRD for 3.5 means that third-party publishers were free to develop those rules. However, I suspect most spell design systems will boil down to some variant of either Just eyeball it! like the official rules and Ultimate Magic or Try this point system then eyeball it! like Chaos and Wizard II.