Well by definition complete deletion means that you are telling the software to eliminate all the sites with gaps so as to consider them not to be informative at all. this is useful when you know that the shown gaps are the result of technical problems such as the fat that you have sequences of different lengths and therefore the shown gaps are just missing data not real informative gaps.

As in partial deletion the story goes differently. Using this option you decide on a cutoff value and based on that your gaps might or might not be considered as informative sites based on the value you have set.

Hope this helps a little  Diego.

Cheers

one more thing it is best to use the partial deletion option as it lets you not to miss any valuable data (of course with a right cutoff value).

Complete deletion ensures that all analysed sequences are the same length and have complete data (no indels or uncertainties). This is a worthwhile objective if you have a reasonably complete dataset and you are not using indels as informative data. It can also substantially reduce analysis time.

Partial deletion allows you to make use of informative changes at sites where there is missing data or indels for some taxa. If you decide on a high cut-off, e.g. exclude sites with < 90% data, this is similar to complete deletion but avoids the problem of ambiguities in one taxon excluding otherwise informative characters. In this way it allows you to include taxa with missing data.

Another option in some programs is pairwise deletion, where the decision is made in pairwise comparisons among taxa. A character is excluded when it is absent in one of both of two taxa compared. The character is included when it is present in both taxa. This maximises the use of all characters but is not applicable to some analyses.

No deletion is the equivalent for analyses with an explicit substitional model such as ML. This is inefficient and can introduce uncertainty to the results. A little bit of ambiguity greatly increases the resources and time used in analysis. If the missing data has little relevance to the relationships of most interest then those gap characters would be better excluded.

Thank you all of you firstly, the advices are helpful. May I ask two questionions that I am still confused about after reading many related topics.

I used 60 complete amino acid sequences which were translated from genome or from published papers to make alignment for EGFR. And then, I have tried "complete, pairwise, partial (95% site coverage) deletion when I make a evolutionary tree with NJ method. I find the bootstrap values are different a lot, and the clustering of species with a little different but quite simmilar. The bootstrap values (1000 replications)got from them ( high to low) is partial>pairwise>complete, So shall I choose the partial deletion method which get the best bootstrap values?

In my opinion, the sequence I used which were translated from genome or from published papers were reliable. So I need to use complete deletion method? But the it's inconsistent with the bootstrap values.

I have seen some adviced that pairwise deletion is better, but some said complete deletion is better, how can I understand it? Thank you for your help!