FASTQ Quality Score Calculator — Phred Q Score, Error Rate, Q20/Q30

The fourth line of a FASTQ file — that string of cryptic symbols — encodes quality scores. This tool does two things: converts between Q scores and error rates; parses a quality string into mean quality, ≥Q20/Q30 fractions, and expected error count.

A Q score is just a log-transformed error rate

The PHRED quality score has a one-line definition:

Q = −10 × log₁₀(Pe)      Pe is the probability of a base-calling error

So every 10-point increase in Q drops the error rate by one order of magnitude:

Metrics like “90% of bases ≥Q30” are common because Q30 corresponds to a 0.1% error rate, which is low enough for most applications.

Mean quality cannot be computed by averaging Q scores directly

This is the most common mistake — and the error is not small.

Q scores are logarithms. The arithmetic mean of logarithms has no physical meaning. To compute the mean quality of a read, you must convert each Q to an error rate, average the error rates, then convert back:

Correct: Q_mean = −10 × log₁₀( mean(Pe) )
Wrong:   Q_mean = mean(Q)

How large is the difference? Take a read that is half Q40 and half Q20 (50 bases each):

A difference of 7 Q units. Low-quality bases have very high error rates that dominate when averaging in error-rate space, while in the arithmetic mean of Q values each base counts equally.

More extreme: [Q40, Q40, Q40, Q10] gives an arithmetic mean of Q32.5, while the correct method yields only Q16.0 — a single Q10 base pulls the whole read’s mean quality down to Q16.

This is not pedantry: if you filter reads using the arithmetic mean, you will pass reads that actually carry many errors into downstream analyses.

Three mutually incompatible FASTQ encodings

Quality scores are stored as single ASCII characters, but three encodings have appeared historically, and they cannot be reliably distinguished automatically:

Note the middle row: early Solexa used Solexa scores, not PHRED scores: Q = −10 × log₁₀(Pe / (1 − Pe)), which can be negative. The two score types converge at high quality but diverge at low quality — PHRED 10 corresponds to Solexa 9.5, while PHRED 2 corresponds to Solexa −2.3.

Data you encounter today is almost always Phred+33. Phred+64 only appears in data from before 2011.

About automatic encoding detection

This tool attempts to detect the encoding, but only reports a conclusion when it can be certain:

• A character with ASCII code below 64 → must be Phred+33;
• The highest character is above J (74) → inferred as Phred+64, because Illumina 1.8+ caps Q values at about 41 (corresponding to J);
• Neither condition met → the tool tells you it cannot determine the encoding; please specify manually.

No guessing. Short reads or high-quality reads may fall entirely within the overlapping region of both encodings. A wrong guess shifts every Q value by 31 — worse than giving no answer at all.

Expected error count is more useful than mean quality

The “expected number of erroneous bases” in the output is the direct sum of all Pe values. Its meaning is straightforward: how many bases in this read are expected to be wrong.

Filtering reads by expected error count is more principled than by mean quality — mean quality is poorly comparable across reads of different lengths, whereas “expected number of wrong bases” is on the same scale at any read length. Long-read QC tools use this metric for exactly this reason.

Data sources

Encoding offsets and score ranges are taken from the Biopython Bio.SeqIO.QualityIO module source code, which follows the defining paper for the FASTQ format: Cock PJA, Fields CJ, Goto N, Heuer ML, Rice PM (2010) The Sanger FASTQ file format for sequences with quality scores, and the Solexa/Illumina FASTQ variants. Nucleic Acids Research 38(6):1767–1771, DOI 10.1093/nar/gkp1137.

PHRED–Solexa conversions have been verified point-by-point against Biopython’s solexa_quality_from_phred().

Related tools

To calculate sequencing depth and data volume, see Sequencing Coverage Calculator; for basic sequence statistics, see Sequence Basic Statistics.