Where can I find a dataset with all (or the most important) words from a language? For example, I need a dictionary for the Romanian language.
In the first stage just the words, the frequency would be useful also, nothing else.
It seems that you're looking for a lexicon, which normally contain words with their inflections, derivatives, frequencies, etc, derived from a corpus of text in a particular language.
The only thing that I could find for Romanian is the Romanian WordForm Lexicon, which contains lexical data extracted from the Romanian Balanced Corpus, and it has slightly restricted use. It seems that you need to contact Dan Tufis, who is the maintainer. Information about the lexicon can be found here, including contact information for Dan Tufis, the structure of the lexicon, how to cite the lexicon, etc.
You may also be interested in the Romanian WordNet, which contains sets of cognitive synonyms; it is a derivative work of Princeton's original WordNet project for English, and falls under the same jurisdiction as the Romanian WordForm Lexicon.
Unfortunately I don't know what is required to use MetaShare (which hosts/licenses all of these documents); it is part of MetaNet, the European Multilingual Technology Alliance.
Specifically looking for Romanian, there is babelnet 2.5.1 (a multilingual wordnet) available under an Open licence from http://babelnet.org/. (Open Multilingual Wordnet has no Romanian section yet.)
A free corpus containg a good wealth of Romanian is europarl available from many places.
For a dictionary, consider wiktionary (both ro.wiktionary.org and the Romanian entries from en.wiktionary.org).
Anything else I am aware of may be free of charge for academic research, but not Open Data (e.g. the Leipzig Wortschatz corpora, http://corpora.informatik.uni-leipzig.de/)