You need to start by finding a standard deviation for weight, amount of wool, or whatever you are measuring. Then sample size must be calculated for whatever you see as the  meaningful minimum difference. Even then the  power analysis (sample size) must be calculated for the particular test you are using. Sometimes this requires a pilot study of about 25 samples before the actual experiment (using the control standard deviation is fine). But 'historical' records for that kind of sheep in that environment are also usable.  Statistical packages and the internet might provide the rest as long as you avoid the disastrous and discredited 'small, medium, or large" stuff.

Eljelani -

Lewis has a good answer if you are applying a hypothesis test.  Note that such sample size considerations show the importance of having a type II probability error analysis or equivalent, not just a p-value. 

You don't have to be doing an hypothesis test though.  The same start is important, the population or subpopulation/stratum standard deviation.  These standard deviations are fixed, but must be estimated.  Then you choose a sample size to go with said estimated standard deviation, to arrive at a desirable estimated standard error (or less), for say an estimated mean or a total, or for yes/no data, an estimated proportion. 

I think that Blair, E. and Blair, J(2015), Applied Survey Sampling, Sage Publications, demonstrates both this method and that for power analysis noted by Lewis, and you can find many other resources.  You might want to see a book such as Cochran, W.G(1977), Sampling Techniques, 3rd ed., John Wiley & Sons.  In Cochran there is a good chapter on estimating sample size requirements for means, totals, and proportions when using simple random sampling, followed by information for more complex designs.  If you have auxiliary/regressor data, you may be able to use model-based or model-assisted design-based methods, but that may be beyond the scope of your question.  Regardless, the basic building block is an estimate of a population or subpopulation standard deviation. 

Confidence intervals can then be used if you like, which are often more practically interpretable than hypothesis tests.  One example may be that of confidence intervals around differences in means between two groups. 

Be careful of online sample size 'calculators.'  They are usually only for proportions, simple random sampling, and worst case (p=q=0.5, which avoids estimation of standard deviation but generally results in an overestimate of sample size needs), and usually ignores a finite population correction (fpc) factor, which may result in estimating a sample size that is bigger than the population size.  Regardless, as noted, it assumes you are only looking at yes/no data.

For more than one variable of interest, you may have to compromise. You might just want to be certain you have enough good quality data for your most important questions. 

For other kinds of data there may be other literature but I imagine that the same principles apply: you reduce bias as much as possible, and then see how large a sample of good quality data is needed to see through the variability. 

Cheers - Jim

Because you have a relatively small population size (especially if you could stratify), though the sample size needs depend upon standard deviation(s), for any likely sample size, N-n is liable to be enough below N that you will likely want to consider finite population correction.   (An hypothesis test would implicitly assume an infinite population, whereas I think you want to consider a finite population here.) 

For a definition of a finite population correction factor, you could see the following Sage encyclopedia entry:

 https://www.researchgate.net/publication/262970761_FINITE_POPULATION_CORRECTION_fpc_FACTOR

Article FINITE POPULATION CORRECTION (fpc) FACTOR

thanks Lewis and James for your unlimited answer. I am not measuring the weight actually. in fact, I am screening the prevalence of parasitic infection Cryptosporidium spp in those animals applying longitudinal study (monthly screening of the same group for 2 years period) by randomly sampling those sheep. I am not sure if this kind of screening requires me to achieve certain sample size every months?

If you are following the same individual sheep, perhaps you should look up "panel survey."  Yes there are sample size guidelines.  Not really my area though. 

If you are considering a different random sample each month, you could adjust the sample size based on standard deviation and standard error results from the previous month. 

If this is a yes/no question, as noted above, use sample size for estimating proportion, using p & q found last month as a guess for new month.  Think you need fpc.  See above. See pages 75 and 76 in Cochran(1977) above if simple random sample.  (I checked an internet available explanation that I expected would be good, but it looks to me like they have an error.  Best to see a textbook.)  

 Thanks James. your information is really helpful. I am actually targeting the same group of sheep every months. Targeting group of 500 sheep, every months need to collected random samples from that group for a period of 2 years.