Spoken language is much older than written language. Many beginning conlangers try to build their first languages out of written words without paying too much attention to what they will sound like, but this approach, while usable, is really the wrong way 'round. Before any written language existed, all the rules that govern how languages work — the rules of grammar — were already in place; what they determined was how language was built from patterns of sound.
What sounds will you have?Edit
The first step is to decide on what sounds to use in your language. No languages use precisely the same set of sounds; for example, the sound "th" in English does not occur in most other languages (although it does in some, such as Swahili, Greek, and European Spanish); conversely, many languages contain sounds that English does not use at all.
You probably already know that sounds are classed as consonants and vowels. In deciding on what sounds to use in your language, it often helps to consider these two categories separately. A chart of consonant and vowel sounds is available in an appendix. For now, don't worry about the symbols used; just enjoy hearing some of the exotic sounds you might want to add to your language! But don't get too exotic; make sure that you can pronounce all the sounds you're going to use, and that you can tell them apart easily. (There will be plenty of time for building conlangs that you can't pronounce yourself after you've built ones you can — like a juggler practicing with eyes open before trying it blindfolded!) Another guideline is: don't add more sounds than you will actually need. If you want to use lots of non-English sounds, maybe think about deleting some sounds that English does use as well.
But regardless of what sounds you ultimately decide to use or not use, remember that it is important not to throw in every sound you come across; limit yourself to a system of sounds, rather than every sound that exists. If you find yourself with 50 sounds without a good reason why, you have probably gone overboard. The sounds should also be similar to one another in some way; avoid having "freak" sounds that have no relatives. For example, if you have an aspirated stop (a hard consonant like "t", with a puff of air after it), try to have more than one (but not every aspirated stop).
Choose a convenient way to spell out the sounds you've chosen, so that you can easily write about them, and be sure to write down a clear explanation of just what your spelling rules are. It's a good idea to imitate standard descriptions of your sounds, rather than devising your own; that's the best way to be sure it's clear — even to you, some time later — just what sounds you intended. In the Intermediate level of this wikibook, we'll cover the computer-friendly notation CXS (short for Conlang X-SAMPA), and its big brother IPA (the International Phonetic Alphabet), which are ways to spell out sounds without depending on the spelling rules of any particular language.
What sounds can go where?Edit
When you have your sounds, you will have to work on ways of putting them together. Languages differ greatly in this respect: for example, English usually requires every syllable to contain a vowel; even words with no vowel written, such as the mathematical term "nth", are pronounced with one — "enth". Not all languages follow this rule; some, on the other hand, require every syllable to end with a vowel. If you are going to allow consonants to come after vowels at the end of syllables, you might only allow one consonant, or you might allow a cluster (as English does). Some sounds might be prohibited from certain positions; for instance, the "ng" sound in English can never begin a word, but in Tagalog it can. English allows many clusters of consonants, but prohibits many others: a word may begin str, but not zdr. You can make your own decisions on what clusters to allow; Russian and Polish, for example, have no problem with clusters like zdr.
What sounds do I want?Edit
The types of sound used in a language (especially a themed language, such as Elvish) depends on the theme itself. For example, an Elvish language would be very liquid-heavy with many approximants while Orcish would have fewer gliding sounds and more gruff chokes. Keep in mind that there are two kinds of sounds: voiced and unvoiced. Feel free to pick and choose, but remember that if your people can make a voiced sound, they can most likely make a voiceless one, and vice versa.