F-Score Explained: How to Interpret It

Updated 11 November 2026 · Standard formula, shown below

This guide explains how to interpret the F-score (also called F1-score), a single number that summarizes how well a classification model balances precision and recall. You’ll learn what the score means, how to read its value, and how to use it to compare models—no math background beyond basic percentages is needed.

What You’ll Need

A classification model’s confusion matrix (true positives, false positives, false negatives) or its precision and recall values.
Optional: a calculator to compute the F-score manually if you don’t have software.

Step-by-Step

Understand the two components. Precision = true positives / (true positives + false positives). It answers: “Of all items the model flagged as positive, how many were actually positive?” Recall = true positives / (true positives + false negatives). It answers: “Of all actual positive items, how many did the model catch?”
Know the F1 formula. F1 = 2 × (precision × recall) / (precision + recall). This is the harmonic mean, which penalizes a large gap between precision and recall more than a simple average would.
Compute precision and recall first. Use your confusion matrix. For example, if true positives = 80, false positives = 20, false negatives = 40, then precision = 80/(80+20) = 0.80 and recall = 80/(80+40) = 0.667.
Plug into the F1 formula. Using the numbers above: F1 = 2 × (0.80 × 0.667) / (0.80 + 0.667) = 2 × 0.5336 / 1.467 = 1.0672 / 1.467 = 0.727. So the F1 is about 0.73.
Interpret the score on a 0–1 scale. 1.0 means perfect precision and recall (no errors). 0 means either precision or recall is zero. Values above 0.7 are generally considered good for many tasks, but “good” depends on your domain and the cost of false positives vs. false negatives.
Compare F1 across models. When choosing between two classifiers, the one with the higher F1 usually has a better balance of precision and recall. But if your use case cares much more about one metric (e.g., catching all cancers even with false alarms), you might prefer a model with lower F1 but higher recall.
Check for class imbalance. If your dataset has very few positive examples (e.g., 1% fraud), a high F1 can still hide poor performance on the minority class. Always look at the confusion matrix alongside F1.

Common Mistakes

Using accuracy instead of F1. Accuracy can be misleading when classes are imbalanced. For example, a model that always predicts “not fraud” gets 99% accuracy but an F1 of 0 for the fraud class.
Averaging F1 incorrectly. Don’t average per-class F1 scores with a simple mean unless you intend to treat each class equally. Use macro (unweighted) or weighted (by support) averages as appropriate.
Ignoring the threshold. F1 depends on the probability threshold you use to decide “positive.” Changing the threshold changes precision and recall, and thus F1. Always report the threshold.
Rounding too early. If you round precision and recall to 2 decimals before computing F1, you can get a slightly different result. Keep full precision until the final step.

Tips

Use F1 when you need a single number. It’s great for quick model comparisons in presentations or automated hyperparameter tuning.
Look at the macro vs. weighted F1. Macro F1 treats all classes equally; weighted F1 accounts for class size. Choose based on whether rare classes matter more.
Plot precision-recall curves. The F1 is the harmonic mean at a specific threshold. A precision-recall curve shows how F1 changes across thresholds, helping you pick the best trade-off.
Consider F-beta for different weights. If recall is twice as important as precision, use F2-score (beta=2). The formula is (1+β²) × (precision × recall) / (β² × precision + recall).

Troubleshooting

F1 is 0 or undefined. This happens when precision and recall are both 0 (no true positives). Check your model’s predictions—it may be predicting only one class. Re-train or adjust the threshold.
F1 seems too high. If your dataset is highly imbalanced and you’re using a simple average, the F1 for the majority class can dominate. Use the F1 for the minority class specifically.
F1 differs between libraries. Some tools compute macro or weighted averages by default. Read the documentation to know which version you’re getting.
Precision and recall are both high but F1 is lower than expected. That’s mathematically impossible—F1 is always between the two. Re-check your numbers; you might have mixed up true positives and false positives.

Frequently Asked Questions

What does an F1 score of 0.5 mean?

It means the harmonic mean of precision and recall is 0.5. For instance, precision could be 0.5 and recall 0.5, or precision 0.33 and recall 1.0. It’s generally considered a weak model, but context matters.

Is F1 always between 0 and 1?

Yes, because precision and recall are both between 0 and 1, their harmonic mean is also between 0 and 1. A value of 1 is perfect, 0 means no true positives.

Should I use F1 or accuracy?

Use F1 when classes are imbalanced or when false positives and false negatives have different costs. Accuracy is fine when classes are balanced and errors are equally costly.

What is the difference between F1 and F-beta?

F1 gives equal weight to precision and recall. F-beta lets you weight recall higher (beta > 1) or precision higher (beta < 1). For example, F2 weights recall twice as much as precision.

Can F1 be used for multi-class problems?

Yes, but you need to compute F1 per class and then average. Use macro (simple average) or weighted (by number of true instances) depending on your goal.

How do I interpret F1 in a real-world example?

In spam detection, if F1 = 0.85, it means the model balances catching spam (recall) and not flagging legitimate emails (precision) well. A higher F1 suggests fewer missed spams and fewer false alarms.

Method: the standard formula for this calculation, rounded as each result states. See how the calculators work or report an error.